Source code for wbia.algo.hots.chip_match

# -*- coding: utf-8 -*-
"""
python -m utool.util_inspect check_module_usage --pat="chip_match.py"
"""
import logging
import copy
import numpy as np
import utool as ut
import vtool as vt
from os.path import join
from operator import xor
from wbia.algo.hots import hstypes
from wbia.algo.hots import old_chip_match
from wbia.algo.hots import scoring
from wbia.algo.hots import name_scoring
from wbia.algo.hots import _pipeline_helpers as plh  # NOQA

print, rrr, profile = ut.inject2(__name__)
logger = logging.getLogger('wbia')


[docs]class NeedRecomputeError(Exception): pass
DEBUG_CHIPMATCH = False MAX_FNAME_LEN = 80 if ut.WIN32 else 200 TRUNCATE_UUIDS = ut.get_argflag(('--truncate-uuids', '--trunc-uuids'))
[docs]def safeop(op_, xs, *args, **kwargs): return None if xs is None else op_(xs, *args, **kwargs)
[docs]def filtnorm_op(filtnorm_, op_, *args, **kwargs): return ( None if filtnorm_ is None else [safeop(op_, xs, *args, **kwargs) for xs in filtnorm_] )
[docs]def extend_scores(vals, num): if vals is None: return None return np.append(vals, np.full(num, -np.inf))
[docs]def extend_nplists_(x_list, num, shape, dtype): return x_list + ([np.empty(shape, dtype=dtype)] * num)
[docs]def extend_pylist_(x_list, num, val): return x_list + ([None] * num)
[docs]def extend_nplists(x_list, num, shape, dtype): return safeop(extend_nplists_, x_list, num, shape, dtype)
[docs]def extend_pylist(x_list, num, val): return safeop(extend_pylist_, x_list, num, val)
[docs]def convert_numpy_lists(arr_list, dtype, dims=None): new_arrs = [np.array(arr, dtype=dtype) for arr in arr_list] if dims is not None: new_arrs = [vt.atleast_nd(arr, dims) for arr in new_arrs] return new_arrs
[docs]def safecast_numpy_lists(arr_list, dtype=None, dims=None): if arr_list is None: new_arrs = None else: new_arrs = [np.array(arr, dtype=dtype) for arr in arr_list] if dims is not None: new_arrs = [vt.ensure_shape(arr, dims) for arr in new_arrs] return new_arrs
[docs]def aslist(arr): if isinstance(arr, np.ndarray): return arr.tolist() else: return arr
[docs]def convert_numpy(arr, dtype): return np.array(ut.replace_nones(arr, np.nan), dtype=dtype)
[docs]def check_arrs_eq(arr1, arr2): if arr1 is None and arr2 is None: return True elif isinstance(arr1, np.ndarray) and isinstance(arr2, np.ndarray): return np.all(arr1 == arr2) elif len(arr1) != len(arr2): return False elif any(len(x) != len(y) for x, y in zip(arr1, arr2)): return False elif all(np.all(x == y) for x, y in zip(arr1, arr2)): return True else: return False
[docs]def safe_check_lens_eq(arr1, arr2, msg=None): """ Check if it is safe to check if two arrays are equal safe_check_lens_eq(None, 1) safe_check_lens_eq([3], [2, 4]) """ if msg is None: msg = 'outer lengths do not correspond' if arr1 is None or arr2 is None: return True else: assert len(arr1) == len(arr2), msg + '(%r != %r)' % (len(arr1), len(arr2))
[docs]def safe_check_nested_lens_eq(arr1, arr2): """ Check if it is safe to check if two arrays are equal (nested) safe_check_nested_lens_eq(None, 1) safe_check_nested_lens_eq([[3, 4]], [[2, 4]]) safe_check_nested_lens_eq([[1, 2, 3], [1, 2]], [[1, 2, 3], [1, 2]]) safe_check_nested_lens_eq([[1, 2, 3], [1, 2]], [[1, 2, 3], [1]]) """ if arr1 is None or arr2 is None: return True else: safe_check_lens_eq(arr1, arr2, 'outer lengths do not correspond') for count, (x, y) in enumerate(zip(arr1, arr2)): assert len(x) == len( y ), 'inner lengths at position=%r do not correspond (%r != %r)' % ( count, len(x), len(y), )
def _assert_eq_len(list1_, list2_): if list1_ is not None: ut.assert_eq_len(list1_, list2_)
[docs]def prepare_dict_uuids(class_dict, ibs): """ Hacks to ensure proper uuid conversion """ class_dict = class_dict.copy() if 'qaid' not in class_dict and 'qannot_uuid' in class_dict: class_dict['qaid'] = ibs.get_annot_aids_from_uuid(class_dict['qannot_uuid']) if 'daid_list' not in class_dict and 'dannot_uuid_list' in class_dict: class_dict['daid_list'] = ibs.get_annot_aids_from_uuid( class_dict['dannot_uuid_list'] ) if 'dnid_list' not in class_dict and 'dannot_uuid_list' in class_dict: daid_list = class_dict['daid_list'] dnid_list = ibs.get_name_rowids_from_text(class_dict['dname_list']) # if anything is unknown need to set to be negative daid check_set = set([None, ibs.const.UNKNOWN_NAME_ROWID]) dnid_list = [ -daid if dnid in check_set else dnid for daid, dnid in zip(daid_list, dnid_list) ] class_dict['dnid_list'] = dnid_list if 'qnid' not in class_dict and 'qname' in class_dict: qnid = ibs.get_name_rowids_from_text(class_dict['qname']) # if anything is unknown need to set to be negative daid qaid = class_dict['qaid'] qnid = -qaid if qnid == ibs.const.UNKNOWN_NAME_ROWID else qnid class_dict['qnid'] = qnid if 'unique_nids' not in class_dict and 'unique_name_list' in class_dict: # FIXME: there is no notion of which names belong to this nid # unique_nids = ibs.get_name_rowids_from_text(class_dict['unique_name_list']) # class_dict['unique_nids'] = unique_nids dnid_list = class_dict['dnid_list'] # This will probably work... for the short term unique_nids_, name_groupxs_ = vt.group_indices(np.array(dnid_list)) class_dict['unique_nids'] = unique_nids_ return class_dict
class _ChipMatchVisualization(object): """ Abstract class containing the visualization function for ChipMatch """ def show_single_namematch( cm, qreq_, dnid=None, rank=None, fnum=None, pnum=None, homog=ut.get_argflag('--homog'), **kwargs ): r""" CommandLine: python -m wbia --tf ChipMatch.show_single_namematch --show python -m wbia --tf ChipMatch.show_single_namematch --show --qaid 1 python -m wbia --tf ChipMatch.show_single_namematch --show --qaid 1 \ --dpath figures --save ~/latex/crall-candidacy-2015/figures/namematch.jpg python -m wbia --tf _ChipMatchVisualization.show_single_namematch --show --rank=0 --qaid=1 --save rank0.jpg python -m wbia --tf _ChipMatchVisualization.show_single_namematch --show --rank=1 --qaid=1 --save rank1.jpg python -m wbia --tf _ChipMatchVisualization.show_single_namematch --show --rank=2 --qaid=1 --save rank2.jpg Example: >>> # xdoctest: +REQUIRES(--slow) >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', default_qaids=[18]) >>> if True: >>> import matplotlib as mpl >>> from wbia.scripts.thesis import TMP_RC >>> mpl.rcParams.update(TMP_RC) >>> from wbia.viz import viz_matches >>> defaultkw = dict(ut.recursive_parse_kwargs(viz_matches.show_name_matches)) >>> kwargs = ut.argparse_dict(defaultkw, only_specified=True) >>> kwargs.pop('qaid', None) >>> _nid = ut.get_argval('--dnid', default=cm.qnid) >>> rank = ut.get_argval('--rank', default=None) >>> dnid = None if rank is not None else _nid >>> cm.show_single_namematch(qreq_, dnid=dnid, rank=rank, **kwargs) >>> ut.quit_if_noshow() >>> ut.show_if_requested() """ from wbia.viz import viz_matches assert bool(dnid is None) != bool(rank is None), 'must choose one' if dnid is None: dnid = cm.get_rank_name(rank) qaid = cm.qaid if cm.nid2_nidx is None: raise AssertionError('cm.nid2_nidx has not been evaluated yet') # <GET NAME GROUPXS> try: nidx = cm.nid2_nidx[dnid] # if nidx == 144: # raise except KeyError: # def extend(): # pass # cm.daid_list # cm.print_inspect_str(qreq_) # cm_orig = cm # NOQA # cm_orig.assert_self(qreq_) # other_aids = qreq_.daids # Hack to get rid of key error logger.info('CHIP HAS NO GROUND TRUTH MATCHES') cm.assert_self(verbose=False) cm2 = cm.extend_results(qreq_) cm2.assert_self(verbose=False) cm = cm2 nidx = cm.nid2_nidx[dnid] # raise groupxs = cm.name_groupxs[nidx] daids = vt.take2(cm.daid_list, groupxs) dnids = vt.take2(cm.dnid_list, groupxs) assert np.all(dnid == dnids), 'inconsistent naming, dnid=%r, dnids=%r' % ( dnid, dnids, ) groupxs = groupxs.compress(daids != cm.qaid) # </GET NAME GROUPXS> # sort annots in this name by the chip score # HACK USE cm.annot_score_list group_sortx = cm.annot_score_list.take(groupxs).argsort()[::-1] sorted_groupxs = groupxs.take(group_sortx) # get the info for this name name_fm_list = ut.take(cm.fm_list, sorted_groupxs) REMOVE_EMPTY_MATCHES = len(sorted_groupxs) > 3 REMOVE_EMPTY_MATCHES = True if REMOVE_EMPTY_MATCHES: isvalid_list = np.array([len(fm) > 0 for fm in name_fm_list]) MAX_MATCHES = 3 isvalid_list = ut.make_at_least_n_items_valid(isvalid_list, MAX_MATCHES) name_fm_list = ut.compress(name_fm_list, isvalid_list) sorted_groupxs = sorted_groupxs.compress(isvalid_list) name_H1_list = ( None if not homog or cm.H_list is None else ut.take(cm.H_list, sorted_groupxs) ) name_fsv_list = ( None if cm.fsv_list is None else ut.take(cm.fsv_list, sorted_groupxs) ) name_fs_list = ( None if name_fsv_list is None else [fsv.prod(axis=1) for fsv in name_fsv_list] ) name_daid_list = ut.take(cm.daid_list, sorted_groupxs) # find features marked as invalid by name scoring featflag_list = name_scoring.get_chipmatch_namescore_nonvoting_feature_flags( cm, qreq_=qreq_ ) name_featflag_list = ut.take(featflag_list, sorted_groupxs) # Get the scores for names and chips name_score = cm.name_score_list[nidx] name_rank = ut.listfind(aslist(cm.name_score_list.argsort()[::-1]), nidx) name_annot_scores = cm.annot_score_list.take(sorted_groupxs) _ = viz_matches.show_name_matches( qreq_.ibs, qaid, name_daid_list, name_fm_list, name_fs_list, name_H1_list, name_featflag_list, name_score=name_score, name_rank=name_rank, name_annot_scores=name_annot_scores, qreq_=qreq_, fnum=fnum, pnum=pnum, **kwargs ) return _ def show_single_annotmatch( cm, qreq_, daid=None, fnum=None, pnum=None, homog=False, aid2=None, **kwargs ): r""" TODO: rename daid to aid2 CommandLine: python -m wbia.algo.hots.chip_match show_single_annotmatch:0 --show python -m wbia.algo.hots.chip_match show_single_annotmatch:1 --show python -m wbia.algo.hots.chip_match show_single_annotmatch --show --qaids=5245 --daids=5161 --db PZ_Master1 Example: >>> # xdoctest: +REQUIRES(--slow) >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_post_sver('PZ_MTEST', qaid_list=[18]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> daid = cm.groundtruth_daids[0] >>> ut.quit_if_noshow() >>> cm.show_single_annotmatch(qreq_, daid) >>> ut.show_if_requested() Example: >>> # xdoctest: +REQUIRES(--slow) >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> # Make sure we can show results against an aid that wasn't matched >>> ibs, qreq_, cm_list = plh.testdata_post_sver('PZ_MTEST', qaid_list=[18]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> daid = ut.setdiff(qreq_.daids, cm.daid_list)[0] >>> ut.quit_if_noshow() >>> cm.show_single_annotmatch(qreq_, daid) >>> ut.show_if_requested() cm.compress_top_feature_matches(num=1) """ from wbia.viz import viz_matches if aid2 is not None: assert daid is None, 'use aid2 instead of daid kwarg' daid = aid2 if daid is None: idx = cm.argsort()[0] daid = cm.daid_list[idx] else: try: idx = cm.daid2_idx[daid] except KeyError: cm = cm.extend_results(qreq_) idx = cm.daid2_idx[daid] fm = cm.fm_list[idx] H1 = None if not homog or cm.H_list is None else cm.H_list[idx] fsv = None if cm.fsv_list is None else cm.fsv_list[idx] fs = None if fsv is None else fsv.prod(axis=1) showkw = dict(fm=fm, fs=fs, H1=H1, fnum=fnum, pnum=pnum, **kwargs) score = None if cm.score_list is None else cm.score_list[idx] viz_matches.show_matches2( qreq_.ibs, cm.qaid, daid, qreq_=qreq_, score=score, **showkw ) def show_ranked_matches(cm, qreq_, clip_top=6, *args, **kwargs): r""" Plots the ranked-list of name/annot matches using matplotlib Args: qreq_ (QueryRequest): query request object with hyper-parameters clip_top (int): (default = 6) Kwargs: fnum, figtitle, plottype, ...more SeeAlso: wbia.viz.viz_matches.show_matches2 wbia.viz.viz_matches.show_name_matches CommandLine: python -m wbia --tf ChipMatch.show_ranked_matches --show --qaid 1 python -m wbia --tf ChipMatch.show_ranked_matches --qaid 86 --colorbar_=False --show python -m wbia --tf ChipMatch.show_ranked_matches:0 --qaid 86 --colorbar_=False --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> from wbia.viz import viz_matches >>> import wbia >>> if True: >>> import matplotlib as mpl >>> from wbia.scripts.thesis import TMP_RC >>> mpl.rcParams.update(TMP_RC) >>> cm_list, qreq_ = wbia.testdata_cmlist('PZ_MTEST', [1]) >>> defaultkw = dict(ut.recursive_parse_kwargs(viz_matches.show_name_matches)) >>> kwargs = ut.argparse_dict(defaultkw, only_specified=True) >>> ut.delete_dict_keys(kwargs, ['qaid']) >>> kwargs['plottype'] = kwargs.get('plottype', 'namematch') >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> clip_top = ut.get_argval('--clip-top', default=3) >>> print('kwargs = %s' % (ut.repr2(kwargs, nl=True),)) >>> cm.show_ranked_matches(qreq_, clip_top, **kwargs) >>> ut.show_if_requested() Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> from wbia.viz import viz_matches >>> defaultkw = dict(ut.recursive_parse_kwargs(viz_matches.show_name_matches)) >>> kwargs = ut.argparse_dict(defaultkw, only_specified=True) >>> kwargs.pop('qaid', None) >>> kwargs['plottype'] = kwargs.get('plottype', 'namematch') >>> ibs, qreq_, cm_list = plh.testdata_post_sver('PZ_MTEST', qaid_list=[1]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> clip_top = ut.get_argval('--clip-top', default=3) >>> print('kwargs = %s' % (ut.repr2(kwargs, nl=True),)) >>> cm.show_ranked_matches(qreq_, clip_top, **kwargs) >>> ut.show_if_requested() """ idx_list = ut.listclip(cm.argsort(), clip_top) cm.show_index_matches(qreq_, idx_list, *args, **kwargs) def show_daids_matches(cm, qreq_, daids, *args, **kwargs): idx_list = ut.dict_take(cm.daid2_idx, daids) cm.show_index_matches(qreq_, idx_list, *args, **kwargs) def show_index_matches( cm, qreq_, idx_list, fnum=None, figtitle=None, plottype='annotmatch', **kwargs ): import wbia.plottool as pt if fnum is None: fnum = pt.next_fnum() nRows, nCols = pt.get_square_row_cols(len(idx_list), fix=False) if ut.get_argflag('--vert'): # HACK nRows, nCols = nCols, nRows next_pnum = pt.make_pnum_nextgen(nRows, nCols) for idx in idx_list: daid = cm.daid_list[idx] pnum = next_pnum() if plottype == 'namematch': dnid = qreq_.ibs.get_annot_nids(daid) cm.show_single_namematch(qreq_, dnid, pnum=pnum, fnum=fnum, **kwargs) elif plottype == 'annotmatch': cm.show_single_annotmatch(qreq_, daid, fnum=fnum, pnum=pnum, **kwargs) # FIXME: score = vt.trytake(cm.score_list, idx) annot_score = vt.trytake(cm.annot_score_list, idx) score_str = ( 'score = %.3f' % (score,) if score is not None else 'score = None' ) annot_score_str = ( 'annot_score = %.3f' % (annot_score,) if annot_score is not None else 'annot_score = None' ) title = score_str + '\n' + annot_score_str pt.set_title(title) else: raise NotImplementedError('Unknown plottype=%r' % (plottype,)) if figtitle is not None: pt.set_figtitle(figtitle) show_matches = show_single_annotmatch # HACK def ishow_single_annotmatch(cm, qreq_, aid2=None, **kwargs): r""" Iteract with a match to an individual annotation (or maybe name?) Args: qreq_ (QueryRequest): query request object with hyper-parameters aid2 (int): annotation id(default = None) CommandLine: python -m wbia.algo.hots.chip_match --exec-ishow_single_annotmatch --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_post_sver('PZ_MTEST', qaid_list=[1]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> aid2 = None >>> result = cm.ishow_single_annotmatch(qreq_, aid2, noupdate=True) >>> print(result) >>> ut.show_if_requested() """ from wbia.viz.interact import interact_matches # NOQA # if aid == 'top': # aid = cm.get_top_aids(ibs) kwshow = { 'mode': 1, } if aid2 is None: aid2 = cm.get_top_aids(ntop=1)[0] logger.info('[cm] ishow_single_annotmatch aids(%s, %s)' % (cm.qaid, aid2)) kwshow.update(**kwargs) try: inter = interact_matches.MatchInteraction( qreq_.ibs, cm, aid2, qreq_=qreq_, **kwshow ) inter.start() return inter except Exception as ex: ut.printex(ex, 'failed in cm.ishow_single_annotmatch', keys=['aid', 'qreq_']) raise # if not kwargs.get('noupdate', False): # import wbia.plottool as pt # pt.update() ishow_match = ishow_single_annotmatch ishow_matches = ishow_single_annotmatch def ishow_analysis(cm, qreq_, **kwargs): """ CommandLine: python -m wbia.algo.hots.chip_match --exec-_ChipMatchVisualization.ishow_analysis --show Example: >>> # xdoctest: +REQUIRES(--slow) >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> qaid = 18 >>> ibs, qreq_, cm_list = plh.testdata_pre_sver('PZ_MTEST', qaid_list=[qaid]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> ut.quit_if_noshow() >>> cm.ishow_analysis(qreq_) >>> ut.show_if_requested() """ from wbia.viz.interact import interact_qres kwshow = { 'show_query': False, 'show_timedelta': True, } kwshow.update(kwargs) return interact_qres.ishow_analysis(qreq_.ibs, cm, qreq_=qreq_, **kwshow) def show_analysis(cm, qreq_, **kwargs): from wbia.viz import viz_qres kwshow = { 'show_query': False, 'show_timedelta': True, } kwshow.update(kwargs) return viz_qres.show_qres_analysis(qreq_.ibs, cm, qreq_=qreq_, **kwshow) def imwrite_single_annotmatch(cm, qreq_, aid, **kwargs): """ CommandLine: python -m wbia.algo.hots.chip_match --exec-ChipMatch.imwrite_single_annotmatch --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> kwargs = {} >>> kwargs['dpi'] = ut.get_argval('--dpi', int, None) >>> kwargs['figsize'] = ut.get_argval('--figsize', list, None) >>> kwargs['fpath'] = ut.get_argval('--fpath', str, None) >>> kwargs['draw_fmatches'] = not ut.get_argflag('--no-fmatches') >>> kwargs['vert'] = ut.get_argflag('--vert') >>> kwargs['draw_border'] = ut.get_argflag('--draw_border') >>> kwargs['saveax'] = ut.get_argflag('--saveax') >>> kwargs['in_image'] = ut.get_argflag('--in-image') >>> kwargs['draw_lbl'] = ut.get_argflag('--no-draw-lbl') >>> print('kwargs = %s' % (ut.repr2(kwargs),)) >>> cm, qreq_ = wbia.testdata_cm() >>> aid = cm.get_top_aids()[0] >>> img_fpath = cm.imwrite_single_annotmatch(qreq_, aid, **kwargs) >>> ut.quit_if_noshow() >>> # show the image dumped to disk >>> ut.startfile(img_fpath, quote=True) >>> ut.show_if_requested() """ import wbia.plottool as pt import matplotlib as mpl # Pop save kwargs from kwargs save_keys = ['dpi', 'figsize', 'saveax', 'fpath', 'fpath_strict', 'verbose'] save_vals = ut.dict_take_pop(kwargs, save_keys, None) savekw = dict(zip(save_keys, save_vals)) fpath = savekw.pop('fpath') if fpath is None and 'fpath_strict' not in savekw: savekw['usetitle'] = True was_interactive = mpl.is_interactive() if was_interactive: mpl.interactive(False) # Make new figure fnum = pt.ensure_fnum(kwargs.pop('fnum', None)) # fig = pt.figure(fnum=fnum, doclf=True, docla=True) fig = pt.plt.figure(fnum) fig.clf() # Draw Matches cm.show_single_annotmatch(qreq_, aid, colorbar_=False, fnum=fnum, **kwargs) # if not kwargs.get('notitle', False): # pt.set_figtitle(cm.make_smaller_title()) # Save Figure # Setting fig=fig might make the dpi and figsize code not work img_fpath = pt.save_figure(fpath=fpath, fig=fig, **savekw) pt.plt.close(fig) # Ensure that this figure will not pop up if was_interactive: mpl.interactive(was_interactive) # if False: # ut.startfile(img_fpath) return img_fpath @profile def imwrite_single_annotmatch2(cm, qreq_, aid, fpath, **kwargs): """ users newer rendering based code """ import wbia.plottool as pt import matplotlib as mpl # Pop save kwargs from kwargs save_keys = ['dpi', 'figsize', 'saveax', 'verbose'] save_vals = ut.dict_take_pop(kwargs, save_keys, None) savekw = dict(zip(save_keys, save_vals)) was_interactive = mpl.is_interactive() if was_interactive: mpl.interactive(False) # Make new figure fnum = pt.ensure_fnum(kwargs.pop('fnum', None)) # Create figure --- this takes about 19% - 11% of the time depending on settings fig = pt.plt.figure(fnum) fig.clf() # # Draw Matches --- this takes about 48% - 67% of the time depending on settings # wrapped call to show_matches2 cm.show_single_annotmatch(qreq_, aid, colorbar_=False, fnum=fnum, **kwargs) # Write matplotlib axes to an image axes_extents = pt.extract_axes_extents(fig) assert len(axes_extents) == 1, 'more than one axes' extent = axes_extents[0] # with io.BytesIO() as stream: # This call takes 23% - 15% of the time depending on settings fig.savefig(fpath, bbox_inches=extent, **savekw) # stream.seek(0) # data = np.fromstring(stream.getvalue(), dtype=np.uint8) # image = cv2.imdecode(data, 1) # Ensure that this figure will not pop up pt.plt.close(fig) if was_interactive: mpl.interactive(was_interactive) # return image @profile def render_single_annotmatch(cm, qreq_, aid, **kwargs): """ CommandLine: python -m wbia.algo.hots.chip_match --exec-_ChipMatchVisualization.render_single_annotmatch --show utprof.py -m wbia.algo.hots.chip_match --exec-_ChipMatchVisualization.render_single_annotmatch --show utprof.py -m wbia.algo.hots.chip_match --exec-_ChipMatchVisualization.render_single_annotmatch --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> kwargs = {} >>> kwargs['dpi'] = ut.get_argval('--dpi', int, None) >>> kwargs['draw_fmatches'] = False >>> kwargs['vert'] = False >>> kwargs['show_score'] = False >>> kwargs['show_timedelta'] = False >>> kwargs['draw_border'] = False >>> kwargs['in_image'] = False >>> kwargs['draw_lbl'] = False >>> print('kwargs = %s' % (ut.repr2(kwargs),)) >>> cm, qreq_ = wbia.testdata_cm() >>> aid = cm.get_top_aids()[0] >>> import wbia.plottool as pt >>> tt = ut.tic('render image') >>> img = cm.render_single_annotmatch(qreq_, aid, **kwargs) >>> ut.toc(tt) >>> ut.quit_if_noshow() >>> pt.imshow(img) >>> ut.show_if_requested() """ import io import cv2 import wbia.plottool as pt import matplotlib as mpl # Pop save kwargs from kwargs save_keys = ['dpi', 'figsize', 'saveax', 'verbose'] save_vals = ut.dict_take_pop(kwargs, save_keys, None) savekw = dict(zip(save_keys, save_vals)) was_interactive = mpl.is_interactive() if was_interactive: mpl.interactive(False) # Make new figure fnum = pt.ensure_fnum(kwargs.pop('fnum', None)) # Create figure --- this takes about 19% - 11% of the time depending on settings fig = pt.plt.figure(fnum) fig.clf() # # Draw Matches --- this takes about 48% - 67% of the time depending on settings # wrapped call to show_matches2 cm.show_single_annotmatch(qreq_, aid, colorbar_=False, fnum=fnum, **kwargs) # Write matplotlib axes to an image axes_extents = pt.extract_axes_extents(fig) assert len(axes_extents) == 1, 'more than one axes' extent = axes_extents[0] with io.BytesIO() as stream: # This call takes 23% - 15% of the time depending on settings fig.savefig(stream, bbox_inches=extent, **savekw) stream.seek(0) data = np.fromstring(stream.getvalue(), dtype=np.uint8) image = cv2.imdecode(data, 1) # Ensure that this figure will not pop up pt.plt.close(fig) if was_interactive: mpl.interactive(was_interactive) return image def qt_inspect_gui(cm, ibs, ranks_top=6, qreq_=None, name_scoring=False): r""" Args: ibs (IBEISController): wbia controller object ranks_top (int): (default = 6) qreq_ (QueryRequest): query request object with hyper-parameters(default = None) name_scoring (bool): (default = False) Returns: QueryResult: qres_wgt - object of feature correspondences and scores CommandLine: python -m wbia.algo.hots.chip_match --exec-qt_inspect_gui --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_post_sver('PZ_MTEST', qaid_list=[1]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> ranks_top = 6 >>> name_scoring = False >>> qres_wgt = cm.qt_inspect_gui(ibs, ranks_top, qreq_, name_scoring) >>> ut.quit_if_noshow() >>> import wbia.guitool >>> guitool.qtapp_loop(qwin=qres_wgt) """ logger.info('[cm] qt_inspect_gui') from wbia.gui import inspect_gui from wbia import guitool guitool.ensure_qapp() cm_list = [cm] logger.info('[inspect_matches] make_qres_widget') qres_wgt = inspect_gui.QueryResultsWidget( ibs, cm_list, ranks_top=ranks_top, name_scoring=name_scoring, qreq_=qreq_ ) logger.info('[inspect_matches] show') qres_wgt.show() logger.info('[inspect_matches] raise') qres_wgt.raise_() return qres_wgt class _ChipMatchScorers(object): """ Evaluators evaluate the specific score and add it to a dictionary that can maintain multiple different types of scores. These dicts are: cm.algo_name_scores and cm.algo_annot_scores Cannoizers make a specific type of score cannonical via cm.score_list, cm.name_score_list, and cm.annot_score_list """ # --- Evaluators @profile def evaluate_csum_annot_score(cm, qreq_=None): """ Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.scoring import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_pre_sver('testdb1', qaid_list=[1]) >>> cm = cm_list[0] >>> cm.evaluate_dnids(qreq_) >>> cm.qnid = 1 # Hack for testdb1 names >>> gt_flags = cm.get_groundtruth_flags() >>> cm.evaluate_csum_annot_score(qreq_) >>> annot_score_list = cm.algo_annot_scores['csum'] >>> assert annot_score_list[gt_flags].max() > annot_score_list[~gt_flags].max() >>> assert annot_score_list[gt_flags].max() > 10.0 """ fs_list = cm.get_fsv_prod_list() csum_scores = np.array([np.sum(fs) for fs in fs_list]) cm.algo_annot_scores['csum'] = csum_scores @profile def evaluate_nsum_name_score(cm, qreq_): """Calls name scoring logic""" cm.evaluate_dnids(qreq_) fmech_scores = name_scoring.compute_fmech_score(cm, qreq_=qreq_) try: normsum = qreq_.qparams.normsum if normsum: assert False, 'depricated' except AttributeError: pass # cm.algo_name_scores['fmech'] = fmech_scores cm.algo_name_scores['nsum'] = fmech_scores def evaluate_maxcsum_name_score(cm, qreq_): grouped_csum = vt.apply_grouping(cm.algo_annot_scores['csum'], cm.name_groupxs) maxcsum_scores = np.array([scores.max() for scores in grouped_csum]) cm.algo_name_scores['maxcsum'] = maxcsum_scores def evaluate_sumamech_name_score(cm, qreq_): grouped_csum = vt.apply_grouping(cm.algo_annot_scores['csum'], cm.name_groupxs) sumamech_score_list = np.array([scores.sum() for scores in grouped_csum]) cm.algo_name_scores['sumamech'] = sumamech_score_list # --- Cannonizers @profile def score_annot_csum(cm, qreq_): """ CommandLine: python -m wbia.algo.hots.chip_match --test-score_annot_csum --show python -m wbia.algo.hots.chip_match --test-score_annot_csum --show --qaid 18 Example: >>> # xdoctest: +REQUIRES(--slow) >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_post_sver() >>> cm = cm_list[0] >>> cm.score_annot_csum(qreq_) >>> ut.quit_if_noshow() >>> cm.show_ranked_matches(qreq_, figtitle='score_annot_csum') >>> ut.show_if_requested() """ cm.evaluate_csum_annot_score(qreq_) cm.set_cannonical_annot_score(cm.algo_annot_scores['csum']) @profile def score_name_maxcsum(cm, qreq_): """ This is amech from the thesis """ cm.evaluate_dnids(qreq_) cm.evaluate_csum_annot_score(qreq_) cm.evaluate_maxcsum_name_score(qreq_) cm.set_cannonical_name_score( cm.algo_annot_scores['csum'], cm.algo_name_scores['maxcsum'] ) @profile def score_name_nsum(cm, qreq_): """ This is fmech from the thesis CommandLine: python -m wbia.algo.hots.chip_match --test-score_name_nsum --show --qaid 1 python -m wbia.algo.hots.chip_match --test-score_name_nsum --show --qaid 18 -t default:normsum=True Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> qreq_, args = plh.testdata_pre('end', defaultdb='PZ_MTEST', >>> a=['default'], qaid_override=[18]) >>> cm = args.cm_list_SVER[0] >>> cm.score_name_nsum(qreq_) >>> gt_score = cm.score_list.compress(cm.get_groundtruth_flags()).max() >>> cm.print_csv() >>> top_nid = cm.unique_nids[cm.name_score_list.argmax()] >>> assert cm.get_top_nids()[0] == top_nid, 'bug in alignment' >>> ut.quit_if_noshow() >>> cm.show_ranked_matches(qreq_, figtitle='score_name_nsum') >>> ut.show_if_requested() >>> assert cm.get_top_nids()[0] == cm.qnid, 'is this case truely hard?' """ cm.evaluate_csum_annot_score(qreq_) cm.evaluate_nsum_name_score(qreq_) cm.set_cannonical_name_score( cm.algo_annot_scores['csum'], cm.algo_name_scores['nsum'] ) @profile def score_name_sumamech(cm, qreq_): cm.evaluate_csum_annot_score(qreq_) cm.evaluate_sumamech_name_score(qreq_) cm.set_cannonical_name_score( cm.algo_annot_scores['csum'], cm.algo_name_scores['sumamech'] )
[docs]class MatchBaseIO(object): """"""
[docs] @classmethod def load_from_fpath(cls, fpath, verbose=ut.VERBOSE): state_dict = ut.load_cPkl(fpath, verbose=verbose) self = cls() self.__setstate__(state_dict) return self
[docs] def save_to_fpath(cm, fpath, verbose=ut.VERBOSE): """ CommandLine: python wbia --tf MatchBaseIO.save_to_fpath --verbtest --show Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> qaid = 18 >>> ibs, qreq_, cm_list = plh.testdata_pre_sver('PZ_MTEST', qaid_list=[qaid]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> dpath = ut.get_app_resource_dir('wbia') >>> fpath = join(dpath, 'tmp_chipmatch.cPkl') >>> ut.delete(fpath) >>> cm.save_to_fpath(fpath) >>> cm2 = ChipMatch.load_from_fpath(fpath) >>> assert cm == cm2 >>> ut.quit_if_noshow() >>> cm.ishow_analysis(qreq_) >>> ut.show_if_requested() """ # ut.save_data(fpath, cm.__getstate__(), verbose=verbose) ut.save_cPkl(fpath, cm.__getstate__(), verbose=verbose)
def __getstate__(cm): state_dict = cm.__dict__ return state_dict def __setstate__(cm, state_dict): if 'algo_annot_scores' not in state_dict and 'algo_name_scores' not in state_dict: # Move to new dict algo score interface # This can be removed once we are sure all caches made before this # change have been recomputed or deleted. algo_annot_scores = {key: None for key in cm._special_annot_scores} algo_name_scores = {key: None for key in cm._special_name_scores} algo_annot_scores['csum'] = state_dict['csum_score_list'] algo_name_scores['nsum'] = state_dict['nsum_score_list'] state_dict['algo_annot_scores'] = algo_annot_scores state_dict['algo_name_scores'] = algo_name_scores del state_dict['csum_score_list'] del state_dict['nsum_score_list'] del state_dict['acov_score_list'] del state_dict['ncov_score_list'] del state_dict['maxcsum_score_list'] del state_dict['special_annot_scores'] del state_dict['special_name_scores'] cm.__dict__.update(state_dict)
[docs] def copy(self): cls = self.__class__ out = cls() state_dict = copy.deepcopy(self.__getstate__()) out.__setstate__(state_dict) return out
class _BaseVisualization(object): def show_analysis(cm, qreq_, **kwargs): # HACK FOR ANNOT MATCH (HUMPBACKS) from wbia.viz import viz_qres kwshow = { 'show_query': False, 'show_timedelta': True, } kwshow.update(kwargs) # logger.info('\n') # logger.info("???? HACK SHOW QRES ANALYSIS") return viz_qres.show_qres_analysis(qreq_.ibs, cm, qreq_=qreq_, **kwshow) def ishow_analysis(cm, qreq_, **kwargs): # HACK FOR ANNOT MATCH (HUMPBACKS) from wbia.viz.interact import interact_qres kwshow = { 'show_query': False, 'show_timedelta': True, } kwshow.update(kwargs) return interact_qres.ishow_analysis(qreq_.ibs, cm, qreq_=qreq_, **kwshow) def show_single_namematch( cm, qreq_, dnid, fnum=None, pnum=None, homog=ut.get_argflag('--homog'), **kwargs ): """ HACK FOR ANNOT MATCH """ # HACK FOR ANNOT MATCH (HUMPBACKS) # logger.info('\n') # logger.info("???? HACK SHOW SINGLE NAME MATCH") from wbia.viz import viz_matches qaid = cm.qaid if cm.nid2_nidx is None: raise AssertionError('cm.nid2_nidx has not been evaluated yet') # cm.score_name_nsum(qreq_) # <GET NAME GROUPXS> try: nidx = cm.nid2_nidx[dnid] except KeyError: # cm.print_inspect_str(qreq_) cm_orig = cm # NOQA cm_orig.assert_self(qreq_) # Hack to get rid of key error cm.assert_self(verbose=False) cm2 = cm.extend_results(qreq_) cm2.assert_self(verbose=False) cm = cm2 nidx = cm.nid2_nidx[dnid] # raise # </GET NAME GROUPXS> groupxs = cm.name_groupxs[nidx] daids = vt.take2(cm.daid_list, groupxs) dnids = vt.take2(cm.dnid_list, groupxs) assert np.all(dnid == dnids), 'inconsistent naming, dnid=%r, dnids=%r' % ( dnid, dnids, ) groupxs = groupxs.compress(daids != cm.qaid) # </GET NAME GROUPXS> # sort annots in this name by the chip score group_sortx = cm.annot_score_list.take(groupxs).argsort()[::-1] sorted_groupxs = groupxs.take(group_sortx) # get the info for this name name_daid_list = ut.take(cm.daid_list, sorted_groupxs) # find features marked as invalid by name scoring # Get the scores for names and chips name_score = cm.name_score_list[nidx] name_rank = ut.listfind(aslist(cm.name_score_list.argsort()[::-1]), nidx) name_annot_scores = cm.annot_score_list.take(sorted_groupxs) kwargs = kwargs.copy() # logger.info('kwargs.copy = %r' % (kwargs,)) # draw_fmatches = kwargs.get('draw_fmatches', True) # MEGAHACK TO DEAL WITH OLD EXPLICIT ELLIPSE FEATURES kwargs['draw_fmatches'] = kwargs.get('draw_ell', True) kwargs['show_matches'] = False _ = viz_matches.show_name_matches( qreq_.ibs, qaid, name_daid_list, None, None, None, None, name_score=name_score, name_rank=name_rank, name_annot_scores=name_annot_scores, qreq_=qreq_, fnum=fnum, pnum=pnum, **kwargs ) return _ class _AnnotMatchConvenienceGetter(object): # @property # def algo_annot_scores(cm): # attrs = [score_method + '_score_list' for score_method in cm._special_annot_scores] # algo_annot_scores = ut.ClassAttrDictProxy(cm, cm._special_annot_scores, attrs) # return algo_annot_scores # @property # def algo_name_scores(cm): # attrs = [score_method + '_score_list' for score_method in cm._special_name_scores] # algo_name_scores = ut.ClassAttrDictProxy(cm, cm._special_name_scores, attrs) # return algo_name_scores # ------------------ # Score-Based Result Functions # ------------------ def pandas_annot_info(cm): import pandas as pd data = { 'daid': cm.daid_list, 'dnid': cm.dnid_list, 'score': cm.annot_score_list, 'rank': cm.annot_score_list.argsort()[::-1].argsort(), 'truth': (cm.dnid_list == cm.qnid).astype(np.int), } annot_df = pd.DataFrame(data) annot_df.sort_values(by='rank', inplace=True) annot_df.reset_index(inplace=True, drop=True) return annot_df def pandas_name_info(cm): import pandas as pd data = { 'dnid': cm.unique_nids, 'score': cm.name_score_list, 'rank': cm.name_score_list.argsort()[::-1].argsort(), 'truth': (cm.unique_nids == cm.qnid).astype(np.int), } name_df = pd.DataFrame(data) name_df.sort_values(by='rank', inplace=True) name_df.reset_index(inplace=True, drop=True) return name_df def summarize(cm, qreq_): """ Summarize info about the groundtruth and the best groundfalse. """ # ibs = qreq_.ibs cminfo_dict = dict( # annot props gt_aid=None, gf_aid=None, gt_annot_daid=None, gf_annot_daid=None, gt_annot_rank=None, gf_annot_rank=None, gt_annot_score=None, gf_annot_score=None, # name props gt_name_rank=None, gf_name_rank=None, gt_name_score=None, gf_name_score=None, ) # Name and annot info sorted by rank name_df = cm.pandas_name_info() annot_df = cm.pandas_annot_info() name_df = cm.pandas_name_info() for truth, tstr in [(1, 'gt'), (0, 'gf')]: # Name properties idxs = np.where(name_df['truth'] == truth)[0] if len(idxs) > 0: idx = min(idxs) for prop in ['rank', 'score']: key = '{}_name_{}'.format(tstr, prop) cminfo_dict[key] = name_df[prop].iloc[idx] # else: # if truth == 0 and len(cm.dnid_list) < len(name_df): # # Handle the case where the cm list is not extended # randrank = np.random.randint(len(cm.dnid_list), len(name_df)) # key = '{}_name_{}'.format(tstr, 'rank') # cminfo_dict[key] = randrank # key = '{}_name_{}'.format(tstr, 'score') # cminfo_dict[key] = -np.inf # Annot properties idxs = np.where(annot_df['truth'] == truth)[0] if len(idxs) > 0: idx = min(idxs) for prop in ['rank', 'score', 'daid']: key = '{}_annot_{}'.format(tstr, prop) cminfo_dict[key] = annot_df[prop].iloc[idx] cminfo_dict.update( dict(gt_aid=cminfo_dict['gt_annot_daid'], gf_aid=cminfo_dict['gf_annot_daid']) ) del cminfo_dict['gt_annot_daid'] del cminfo_dict['gf_annot_daid'] # old aliases cminfo_dict.update( dict( gt_rank=cminfo_dict['gt_annot_rank'], gf_rank=cminfo_dict['gf_annot_rank'], gt_raw_score=cminfo_dict['gt_annot_score'], gf_raw_score=cminfo_dict['gf_annot_score'], ) ) return cminfo_dict # def get_ranked_nids_and_aids(cm): # """ Hacky func # Returns: # wbia.algo.hots.name_scoring.NameScoreTup # """ # sortx = cm.name_score_list.argsort()[::-1] # sorted_name_scores = cm.name_score_list.take(sortx, axis=0) # sorted_nids = cm.unique_nids.take(sortx, axis=0) # sorted_groupxs = ut.take(cm.name_groupxs, sortx) # sorted_daids = vt.apply_grouping(cm.daid_list, sorted_groupxs) # sorted_annot_scores = vt.apply_grouping(cm.annot_score_list, sorted_groupxs) # # do subsorting # subsortx_list = [scores.argsort()[::-1] for scores in sorted_annot_scores] # subsorted_daids = vt.ziptake(sorted_daids, subsortx_list) # subsorted_annot_scores = vt.ziptake(sorted_annot_scores, subsortx_list) # nscoretup = name_scoring.NameScoreTup(sorted_nids, sorted_name_scores, # subsorted_daids, # subsorted_annot_scores) # return nscoretup def get_annot_ave_precision(cm): import sklearn.metrics annot_df = cm.pandas_annot_info() y_true = annot_df['truth'].values y_score = annot_df['score'].values avep = sklearn.metrics.average_precision_score(y_true, y_score) logger.info('avep = %r' % (avep,)) return avep def get_name_ave_precision(cm): import sklearn.metrics name_df = cm.pandas_name_info() y_true = name_df['truth'].values y_score = name_df['score'].values avep = sklearn.metrics.average_precision_score(y_true, y_score) logger.info('avep = %r' % (avep,)) return avep def get_top_scores(cm, ntop=None): sortx = cm.score_list.argsort()[::-1] _top_scores = vt.list_take_(cm.score_list, sortx) top_scores = ut.listclip(_top_scores, ntop) return top_scores def get_top_nids(cm, ntop=None): sortx_ = cm.score_list.argsort()[::-1] sortx = sortx_[slice(0, ntop)] top_nids = vt.list_take_(cm.dnid_list, sortx) return top_nids def get_top_aids(cm, ntop=None): sortx = cm.score_list.argsort()[::-1] _top_aids = vt.list_take_(cm.daid_list, sortx) top_aids = ut.listclip(_top_aids, ntop) return top_aids def get_top_truth_aids(cm, ibs, truth, ntop=None, invert=False): """top scoring aids of a certain truth value""" sortx = cm.score_list.argsort()[::-1] _top_aids = vt.list_take_(cm.daid_list, sortx) _top_nids = vt.list_take_(cm.dnid_list, sortx) isunknown_list = _top_nids <= 0 if cm.qnid <= 0: isunknown_list[:] = True truth_list = np.array((cm.qnid == _top_nids), dtype=np.int32) truth_list[isunknown_list] = ibs.const.EVIDENCE_DECISION.UNKNOWN # truth_list = ibs.get_match_truths([cm.qaid] * len(_top_aids), _top_aids) flag_list = truth_list == truth if invert: flag_list = np.logical_not(flag_list) _top_aids = _top_aids.compress(flag_list, axis=0) top_truth_aids = ut.listclip(_top_aids, ntop) return top_truth_aids def get_top_gf_aids(cm, ibs, ntop=None): return cm.get_top_truth_aids(ibs, ibs.const.EVIDENCE_DECISION.NEGATIVE, ntop) def get_top_gt_aids(cm, ibs, ntop=None): return cm.get_top_truth_aids(ibs, ibs.const.EVIDENCE_DECISION.POSITIVE, ntop) # ------------------ # Getter Functions # ------------------ def get_name_scores(cm, dnids): # idx_list = [cm.daid2_idx.get(daid, None) for daid in daids] nidx_list = ut.dict_take(cm.nid2_nidx, dnids, None) score_list = [ None if idx is None else cm.name_score_list[idx] for idx in nidx_list ] return score_list def get_name_ranks(cm, dnids): # score_method=None): score_ranks = cm.name_score_list.argsort()[::-1].argsort() idx_list = ut.dict_take(cm.nid2_nidx, dnids, None) rank_list = [None if idx is None else score_ranks[idx] for idx in idx_list] return rank_list # def get_nid_scores(cm, nid_list): # nidx_list = ut.dict_take(cm.nid2_nidx, nid_list) # name_scores = vt.list_take_(cm.name_score_list, nidx_list) # return name_scores def get_rank_name(cm, rank): sorted_nids, sorted_name_scores = cm.get_ranked_nids() return sorted_nids[rank] def get_ranked_nids(cm): sortx = cm.name_score_list.argsort()[::-1] sorted_name_scores = cm.name_score_list.take(sortx, axis=0) sorted_nids = cm.unique_nids.take(sortx, axis=0) return sorted_nids, sorted_name_scores def get_annot_scores(cm, daids, score_method=None): # TODO: how to specify either annot_score_list or score_list? # score_list = cm.annot_score_list score_list = cm.score_list idx_list = ut.dict_take(cm.daid2_idx, daids, None) score_list = [None if idx is None else score_list[idx] for idx in idx_list] return score_list def get_annot_ranks(cm, daids): # score_method=None): score_ranks = cm.score_list.argsort()[::-1].argsort() idx_list = ut.dict_take(cm.daid2_idx, daids, None) rank_list = [None if idx is None else score_ranks[idx] for idx in idx_list] return rank_list def get_groundtruth_flags(cm): assert cm.dnid_list is not None, 'run cm.evaluate_dnids' gt_flags = cm.dnid_list == cm.qnid return gt_flags def get_groundtruth_daids(cm): gt_flags = cm.get_groundtruth_flags() gt_daids = vt.list_compress_(cm.daid_list, gt_flags) return gt_daids def get_groundfalse_daids(cm): gf_flags = np.logical_not(cm.get_groundtruth_flags()) gf_daids = vt.list_compress_(cm.daid_list, gf_flags) return gf_daids @property def groundtruth_daids(cm): return cm.get_groundtruth_daids() def get_num_matches_list(cm): num_matches_list = list(map(len, cm.fm_list)) return num_matches_list def get_name_shortlist_aids(cm, nNameShortList, nAnnotPerName): """ Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_pre_sver('PZ_MTEST', qaid_list=[18]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> top_daids = cm.get_name_shortlist_aids(5, 2) >>> assert cm.qnid in ibs.get_annot_name_rowids(top_daids) """ top_daids = scoring.get_name_shortlist_aids( cm.daid_list, cm.dnid_list, cm.annot_score_list, cm.name_score_list, cm.nid2_nidx, nNameShortList, nAnnotPerName, ) return top_daids def get_annot_shortlist_aids(cm, num_shortlist): """ Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_pre_sver('PZ_MTEST', qaid_list=[18]) >>> cm = cm_list[0] >>> cm.score_name_nsum(qreq_) >>> top_daids = cm.get_annot_shortlist_aids(5 * 2) >>> assert cm.qnid in ibs.get_annot_name_rowids(top_daids) """ sortx = np.array(cm.annot_score_list).argsort()[::-1] topx = sortx[: min(num_shortlist, len(sortx))] top_daids = cm.daid_list[topx] return top_daids @property def num_daids(cm): return None if cm.daid_list is None else len(cm.daid_list) @property def ranks(cm): sortx = cm.argsort() return sortx.argsort() @property def unique_name_ranks(cm): sortx = cm.name_argsort() return sortx.argsort() def argsort(cm): assert cm.score_list is not None, 'no annot scores computed' sortx = np.argsort(cm.score_list)[::-1] return sortx def name_argsort(cm): assert cm.name_score_list is not None, 'no name scores computed' return np.argsort(cm.name_score_list)[::-1]
[docs]class AnnotMatch( MatchBaseIO, ut.NiceRepr, _BaseVisualization, _AnnotMatchConvenienceGetter ): """ This implements part the match between whole annotations and the other annotaions / names. This does not include algorithm specific feature matches. """ _attr_names = [ 'qaid', 'qnid', 'daid_list', 'dnid_list', 'H_list', 'score_list', 'annot_score_list', 'unique_nids', 'name_score_list', ] _special_annot_scores = [ 'csum', # 'acov', ] # Special name scores _special_name_scores = [ 'nsum', # fmech 'maxcsum', # amech 'sumamech', # amech # 'ncov', ] def __init__(cm, *args, **kwargs): cm.qaid = None cm.qnid = None cm.daid_list = None # This is aligned with daid list, do not confuse with unique_nids cm.dnid_list = None cm.H_list = None cm.score_list = None # standard groupings # TODO: rename unique_nids to indicate it is aligned with name_groupxs # Annot scores cm.annot_score_list = None # Name scores cm.unique_nids = None # belongs to name_groupxs cm.name_score_list = None cm.algo_annot_scores = {key: None for key in cm._special_annot_scores} cm.algo_name_scores = {key: None for key in cm._special_name_scores} # for score_method in cm._special_name_scores: # setattr(cm, score_method + '_score_list', None) # for score_method in cm._special_annot_scores: # setattr(cm, score_method + '_score_list', None) # Re-evaluatables (for convinience only) cm.daid2_idx = None # maps onto cm.daid_list cm.nid2_nidx = None # maps onto cm.unique_nids cm.name_groupxs = None def __nice__(cm): return 'qaid=%s nD=%s' % (cm.qaid, cm.num_daids)
[docs] def initialize( cm, qaid=None, daid_list=None, score_list=None, dnid_list=None, qnid=None, unique_nids=None, name_score_list=None, annot_score_list=None, autoinit=True, ): """ qaid and daid_list are not optional. fm_list and fsv_list are strongly encouraged and will probalby break things if they are not there. """ cm.qaid = qaid cm.daid_list = safeop(np.array, daid_list, dtype=hstypes.INDEX_TYPE) cm.score_list = safeop(np.array, score_list, dtype=hstypes.FLOAT_TYPE) # name info cm.qnid = qnid cm.dnid_list = safeop(np.array, dnid_list, dtype=hstypes.INDEX_TYPE) cm.unique_nids = safeop(np.array, unique_nids, dtype=hstypes.INDEX_TYPE) cm.name_score_list = safeop(np.array, name_score_list, dtype=hstypes.FLOAT_TYPE) cm.annot_score_list = safeop(np.array, annot_score_list, dtype=hstypes.FLOAT_TYPE) if autoinit: cm._update_daid_index() if cm.dnid_list is not None: cm._update_unique_nid_index() if DEBUG_CHIPMATCH: cm.assert_self(verbose=True)
[docs] def to_dict(cm, ibs=None): class_dict = cm.__getstate__() if ibs is not None: assert ibs is not None, 'need ibs to convert uuids' class_dict['dannot_uuid_list'] = ibs.get_annot_uuids(cm.daid_list) class_dict['dname_list'] = ibs.get_name_texts(cm.dnid_list) class_dict['qannot_uuid'] = ibs.get_annot_uuids(cm.qaid) class_dict['qname'] = ibs.get_name_texts(cm.qnid) return class_dict
[docs] @classmethod def from_dict(ChipMatch, class_dict, ibs=None): r""" Convert dict of arguments back to ChipMatch object """ key_list = ut.get_kwargs(ChipMatch.initialize)[0] # HACKY key_list.remove('autoinit') if ut.VERBOSE: other_keys = list(set(class_dict.keys()) - set(key_list)) if len(other_keys) > 0: logger.info( 'Not unserializing extra attributes: %s' % (ut.repr2(other_keys)) ) if ibs is not None: class_dict = prepare_dict_uuids(class_dict, ibs) dict_subset = ut.dict_subset(class_dict, key_list) dict_subset['score_list'] = convert_numpy( dict_subset['score_list'], hstypes.FS_DTYPE ) cm = ChipMatch() cm.initialize(**dict_subset) return cm
def _update_daid_index(cm): """ Rebuilds inverted index from aid to internal index """ cm.daid2_idx = safeop(ut.make_index_lookup, cm.daid_list) def _update_unique_nid_index(cm): """ Rebuilds inverted index from nid to internal (name) index """ # assert cm.unique_nids is not None unique_nids_, name_groupxs_ = vt.group_indices(cm.dnid_list) # assert unique_nids_.dtype == hstypes.INTEGER_TYPE if cm.unique_nids is None: assert cm.name_score_list is None, 'name score is misaligned' cm.unique_nids = unique_nids_ cm.nid2_nidx = ut.make_index_lookup(cm.unique_nids) nidx_list = np.array(ut.dict_take(cm.nid2_nidx, unique_nids_)) inverse_idx_list = nidx_list.argsort() cm.name_groupxs = ut.take(name_groupxs_, inverse_idx_list)
[docs] def evaluate_dnids(cm, qreq_=None, ibs=None): if qreq_ is not None: # cm.qnid = qreq_.qannots.loc([cm.qaid]).nids[0] # dnid_list = qreq_.dannots.loc(cm.daid_list).nids cm.qnid = qreq_.get_qreq_annot_nids(cm.qaid) dnid_list = qreq_.get_qreq_annot_nids(cm.daid_list) # ibs = qreq_.ibs elif ibs is not None: cm.qnid = ibs.get_annot_name_rowids(cm.qaid) dnid_list = ibs.get_annot_name_rowids(cm.daid_list) else: assert False, 'no source of dnids' cm.dnid_list = np.array(dnid_list, dtype=hstypes.INDEX_TYPE) cm._update_unique_nid_index()
# ------------------ # State Modification Functions # ------------------ # Cannonical Setters
[docs] @profile def set_cannonical_annot_score(cm, annot_score_list): cm.annot_score_list = annot_score_list # cm.name_score_list = None cm.score_list = annot_score_list
[docs] @profile def set_cannonical_name_score(cm, annot_score_list, name_score_list): cm.annot_score_list = safeop(np.array, annot_score_list, dtype=hstypes.FLOAT_TYPE) cm.name_score_list = safeop(np.array, name_score_list, dtype=hstypes.FLOAT_TYPE) # align with score_list cm.score_list = name_scoring.align_name_scores_with_annots( cm.annot_score_list, cm.daid_list, cm.daid2_idx, cm.name_groupxs, cm.name_score_list, )
class _ChipMatchConvenienceGetter(object): # ------------------ # Getter Functions # ------------------ def get_flat_fm_info(cm, flags=None): r""" Returns: dict: info_ CommandLine: python -m wbia.algo.hots.chip_match --exec-get_flat_fm_info --show Example: >>> # DISABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_pre_sver( >>> defaultdb='PZ_MTEST', qaid_list=[18]) >>> cm = cm_list[0] >>> info_ = cm.get_flat_fm_info() >>> ut.assert_all_eq(ut.lmap(len, info_.values())) >>> result = ('info_ = %s' % (ut.repr3(info_, precision=2),)) >>> print(result) """ import vtool as vt if flags is None: flags = [True] * len(cm.daid_list) # flags = cm.score_list > 0 # Compress to desired info fsv_list = ut.compress(cm.fsv_list, flags) fm_list = ut.compress(cm.fm_list, flags) daid_list = ut.compress(cm.daid_list, flags) # Flatten on a feature level len_list = [fm.shape[0] for fm in fm_list] info_ = {} nfilt = len(cm.fsv_col_lbls) info_['fsv'] = vt.safe_cat(fsv_list, axis=0, default_shape=(0, nfilt)) info_['fm'] = vt.safe_cat( fm_list, axis=0, default_shape=(0, 2), default_dtype=hstypes.FM_DTYPE ) info_['aid1'] = np.full(sum(len_list), cm.qaid, dtype=hstypes.INDEX_TYPE) info_['aid2'] = vt.safe_cat( [ np.array([daid] * n, dtype=hstypes.INDEX_TYPE) for daid, n in zip(daid_list, len_list) ], default_shape=(0,), default_dtype=hstypes.INDEX_TYPE, ) return info_ def get_num_feat_score_cols(cm): return len(cm.fsv_col_lbls) def get_fsv_prod_list(cm): return [fsv.prod(axis=1) for fsv in cm.fsv_list] def get_annot_fm(cm, daid): idx = ut.dict_take(cm.daid2_idx, daid) fm = ut.take(cm.fm_list, idx) return fm def get_fs_list(cm, colx=None, col=None): assert xor(colx is None, col is None) if col is not None: colx = cm.fsv_col_lbls.index(col) fs_list = [fsv.T[colx].T for fsv in cm.fsv_list] return fs_list @property def qfxs_list(cm): return [fm.T[0] for fm in cm.fm_list] @property def dfxs_list(cm): return [fm.T[1] for fm in cm.fm_list] @property def nfxs_list(cm): nfxs_list = cm.filtnorm_fxs[0] return nfxs_list @property def naids_list(cm): naids_list = cm.filtnorm_aids[0] return naids_list class _ChipMatchDebugger(object): # ------------------ # String Functions # ------------------ def print_inspect_str(cm, qreq_): logger.info(cm.get_inspect_str(qreq_)) def print_rawinfostr(cm): logger.info(cm.get_rawinfostr()) def print_csv(cm, *args, **kwargs): logger.info(cm.get_cvs_str(*args, **kwargs)) def inspect_difference(cm, other, verbose=True): logger.info('Checking difference') raw_infostr1 = cm.get_rawinfostr(colored=False) raw_infostr2 = other.get_rawinfostr(colored=False) difftext = ut.get_textdiff(raw_infostr1, raw_infostr2, num_context_lines=4) if len(difftext) == 0: if verbose: logger.info('no difference') return True else: if verbose: ut.print_difftext(difftext) return False def get_inspect_str(cm, qreq_): r""" Args: qreq_ (QueryRequest): query request object with hyper-parameters Returns: str: varinfo CommandLine: python -m wbia.algo.hots.chip_match --exec-get_inspect_str Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', a='default:dindex=0:10,qindex=0:1', t='best:SV=False') >>> varinfo = cm.get_inspect_str(qreq_) >>> result = ('varinfo = %s' % (str(varinfo),)) >>> print(result) """ # cm.assert_self(qreq_) top_lbls = [' top aids', ' scores', ' ranks'] ibs = qreq_.ibs top_aids = cm.get_top_aids(6) top_scores = cm.get_annot_scores(top_aids) # top_rawscores = np.array(cm.get_aid_scores(top_aids, rawscore=True), dtype=np.float64) top_ranks = np.arange(len(top_aids)) top_list = [top_aids, top_scores, top_ranks] top_lbls += [' isgt'] istrue = ibs.get_match_truths([cm.qaid] * len(top_aids), top_aids) top_list.append(np.array(istrue, dtype=np.int32)) top_lbls = ['top nid'] + top_lbls top_list = [ibs.get_annot_name_rowids(top_aids)] + top_list top_stack = np.vstack(top_list) # top_stack = np.array(top_stack, dtype=object) top_stack = np.array(top_stack, dtype=np.float64) # np.int32) top_str = np.array_str( top_stack, precision=3, suppress_small=True, max_line_width=200 ) top_lbl = '\n'.join(top_lbls) inspect_list = [ 'QueryResult', qreq_.get_cfgstr(), ] if ibs is not None: gt_aids = ut.aslist(cm.get_top_gt_aids(qreq_.ibs)) gt_ranks = cm.get_annot_ranks(gt_aids) gt_scores = cm.get_annot_scores(gt_aids) inspect_list.append('len(cm.daid_list) = %r' % len(cm.daid_list)) inspect_list.append('len(cm.unique_nids) = %r' % len(cm.unique_nids)) inspect_list.append('gt_ranks = %r' % gt_ranks) inspect_list.append('gt_aids = %r' % gt_aids) inspect_list.append('gt_scores = %s' % ut.repr2(gt_scores, precision=6)) inspect_list.extend( [ 'qaid=%r ' % cm.qaid, 'qnid=%r ' % cm.qnid, ut.hz_str(top_lbl, ' ', top_str), # 'num feat matches per annotation stats:', # ut.indent(ut.repr2(nFeatMatch_stats)), # ut.indent(nFeatMatch_stats_str), ] ) inspect_str = '\n'.join(inspect_list) # inspect_str = ut.indent(inspect_str, '[INSPECT] ') return inspect_str def get_rawinfostr(cm, colored=None): r""" Returns: str: varinfo CommandLine: python -m wbia.algo.hots.chip_match get_rawinfostr Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', a='default:dindex=0:10,qindex=0:1', t='best:SV=False') >>> varinfo = cm.get_rawinfostr() >>> result = ('varinfo = %s' % (varinfo,)) >>> print(result) """ def varinfo(varname, onlyrepr=False, canshowrepr=True, cm=cm, varcolor='yellow'): varval = getattr(cm, varname.replace('cm.', '')) varinfo_list = [] print_summary = not onlyrepr and ut.isiterable(varval) show_repr = True show_repr = show_repr or (onlyrepr or not print_summary) symbol = '*' if colored is not False and ut.util_dbg.COLORED_EXCEPTIONS: varname = ut.color_text(varname, varcolor) if show_repr: varval_str = ut.repr2(varval, precision=2) if len(varval_str) > 100: varval_str = '<omitted>' varval_str = ut.truncate_str(varval_str, maxlen=50) varinfo_list += [' * %s = %s' % (varname, varval_str)] symbol = '+' if print_summary: depth = ut.depth_profile(varval) if not show_repr: varinfo_list += [ # ' %s varinfo(%s):' % (symbol, varname,), ' %s %s = <not shown!>' % (symbol, varname) ] varinfo_list += [' len = %r' % (len(varval),)] if depth != len(varval): depth_str = ut.truncate_str(str(depth), maxlen=70) varinfo_list += [' depth = %s' % (depth_str,)] varinfo_list += [' types = %s' % (ut.list_type_profile(varval),)] # varinfo = '\n'.join(ut.align_lines(varinfo_list, '=')) aligned_varinfo_list = ut.align_lines(varinfo_list, '=') varinfo = '\n'.join(aligned_varinfo_list) return varinfo str_list = [] append = str_list.append attr_order = [ 'cm.qaid', 'cm.qnid', 'cm.unique_nids', 'cm.daid_list', 'cm.dnid_list', 'cm.fs_list', 'cm.fm_list', 'cm.fk_list', 'cm.fsv_list', 'cm.fsv_col_lbls', 'cm.filtnorm_aids', 'cm.filtnorm_fxs', 'cm.H_list', 'cm.score_list', 'cm.annot_score_list', 'cm.name_score_list', # 'cm.sumamech_score_list', 'cm.nid2_nidx', 'cm.daid2_idx', ] attrs_ = [attr.replace('cm.', '') for attr in attr_order] unspecified_attrs = sorted(set(cm.__dict__.keys()) - set(attrs_)) append('ChipMatch:') for attr in attr_order: append(varinfo(attr)) for attr in unspecified_attrs: append(varinfo(attr, varcolor='red')) infostr = '\n'.join(str_list) return infostr def get_cvs_str(cm, numtop=6, ibs=None, sort=True): r""" Args: numtop (int): (default = 6) ibs (IBEISController): wbia controller object(default = None) sort (bool): (default = True) Returns: str: csv_str Notes: Very weird that it got a score qaid 6 vs 41 has [72, 79, 0, 17, 6, 60, 15, 36, 63] [72, 79, 0, 17, 6, 60, 15, 36, 63] [72, 79, 0, 17, 6, 60, 15, 36, 63] [0.060, 0.053, 0.0497, 0.040, 0.016, 0, 0, 0, 0] [7, 40, 41, 86, 103, 88, 8, 101, 35] makes very little sense CommandLine: python -m wbia.algo.hots.chip_match --test-get_cvs_str --force-serial Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> ibs, qreq_, cm_list = plh.testdata_post_sver() >>> cm = cm_list[0] >>> numtop = 6 >>> ibs = None >>> sort = True >>> csv_str = cm.get_cvs_str(numtop, ibs, sort) >>> result = ('csv_str = \n%s' % (str(csv_str),)) >>> print(result) """ if not sort or cm.score_list is None: if sort: logger.info('Warning: cm.score_list is None and sort is True') sortx = list(range(len(cm.daid_list))) else: sortx = ut.list_argsort(cm.score_list, reverse=True) if ibs is not None: qnid = ibs.get_annot_nids(cm.qaid) dnid_list = ibs.get_annot_nids(cm.daid_list) else: qnid = cm.qnid dnid_list = cm.dnid_list # Build columns for the csv, filtering out unavailable information column_lbls_ = [ 'daid', 'dnid', 'score', 'num_matches', 'annot_scores', 'fm_depth', 'fsv_depth', ] column_list_ = [ vt.list_take_(cm.daid_list, sortx), None if dnid_list is None else vt.list_take_(dnid_list, sortx), None if cm.score_list is None else vt.list_take_(cm.score_list, sortx), vt.list_take_(cm.get_num_matches_list(), sortx), None if cm.annot_score_list is None else vt.list_take_(cm.annot_score_list, sortx), # None if cm.name_score_list is None else vt.list_take_(cm.name_score_list, sortx), ut.lmap(str, ut.depth_profile(vt.list_take_(cm.fm_list, sortx))), ut.lmap(str, ut.depth_profile(vt.list_take_(cm.fsv_list, sortx))), ] isnone_list = ut.flag_None_items(column_list_) column_lbls = ut.filterfalse_items(column_lbls_, isnone_list) column_list = ut.filterfalse_items(column_list_, isnone_list) # Clip to the top results if numtop is not None: column_list = [ut.listclip(col, numtop) for col in column_list] # hard case for python text parsing # better know about quoted hash symbols header = ut.codeblock( """ # qaid = {qaid} # qnid = {qnid} # fsv_col_lbls = {fsv_col_lbls} """ ).format(qaid=cm.qaid, qnid=qnid, fsv_col_lbls=cm.fsv_col_lbls) csv_str = ut.make_csv_table(column_list, column_lbls, header, comma_repl=';') return csv_str # ------------------ # Testing Functions # ------------------ def assert_self( cm, qreq_=None, ibs=None, strict=False, assert_feats=True, verbose=ut.NOT_QUIET ): # return list1_ is None or len(list1_) == len(list2_) try: assert cm.qaid is not None, 'must have qaid' assert cm.daid_list is not None, 'must give daids' _assert_eq_len(cm.fm_list, cm.daid_list) _assert_eq_len(cm.fsv_list, cm.daid_list) _assert_eq_len(cm.fk_list, cm.daid_list) _assert_eq_len(cm.H_list, cm.daid_list) _assert_eq_len(cm.score_list, cm.daid_list) _assert_eq_len(cm.dnid_list, cm.daid_list) except AssertionError: cm.print_rawinfostr() raise if ibs is None and qreq_ is not None: ibs = qreq_.ibs testlog = TestLogger(verbose=verbose) with testlog.context('lookup score by daid'): if cm.score_list is None: testlog.skip_test() else: daids = cm.get_top_aids() scores = cm.get_top_scores() scores_ = cm.get_annot_scores(daids) if not np.all(scores == scores_): testlog.log_failed('score mappings are NOT ok') with testlog.context('dnid_list = name(daid_list)'): if strict or ibs is not None and cm.dnid_list is not None: nid_list = ibs.get_annot_name_rowids(cm.daid_list) if not np.all(cm.dnid_list == nid_list): testlog.log_failed('annot aligned nids are NOT ok') else: testlog.skip_test() if strict or cm.unique_nids is not None: with testlog.context('unique nid mapping'): assert cm.nid2_nidx is not None, 'name mappings are not built' nidx_list = ut.dict_take(cm.nid2_nidx, cm.unique_nids) assert nidx_list == list(range(len(nidx_list))) assert np.all(cm.unique_nids[nidx_list] == cm.unique_nids) with testlog.context('allsame(grouped(dnid_list))'): grouped_nids = vt.apply_grouping(cm.dnid_list, cm.name_groupxs) for nids in grouped_nids: if not ut.allsame(nids): testlog.log_failed( 'internal dnid name grouping is NOT consistent' ) with testlog.context('allsame(name(grouped(daid_list)))'): if ibs is None: testlog.skip_test() else: # this might fail if this result is old and the names have changed grouped_aids = vt.apply_grouping(cm.daid_list, cm.name_groupxs) grouped_mapped_nids = ibs.unflat_map( ibs.get_annot_name_rowids, grouped_aids ) for nids in grouped_mapped_nids: if not ut.allsame(nids): testlog.log_failed( 'internal daid name grouping is NOT consistent' ) with testlog.context('dnid_list - unique_nid alignment'): grouped_nids = vt.apply_grouping(cm.dnid_list, cm.name_groupxs) for nids, nid in zip(grouped_nids, cm.unique_nids): if not np.all(nids == nid): testlog.log_failed( 'cm.unique_nids is NOT aligned with ' 'vt.apply_grouping(cm.dnid_list, cm.name_groupxs). ' ' nids=%r, nid=%r' % (nids, nid) ) break if ibs is not None: testlog.start_test('daid_list - unique_nid alignment') for nids, nid in zip(grouped_mapped_nids, cm.unique_nids): if not np.all(nids == nid): testlog.log_failed( 'cm.unique_nids is NOT aligned with ' 'vt.apply_grouping(name(cm.daid_list), cm.name_groupxs). ' ' name(aids)=%r, nid=%r' % (nids, nid) ) break testlog.end_test() assert len(testlog.failed_list) == 0, '\n'.join(testlog.failed_list) testlog.log_passed('lengths are ok') try: with testlog.context('check fm_shape'): if cm.fm_list is None: testlog.skip_test() else: assert ut.list_all_eq_to( [fm.shape[1] for fm in cm.fm_list], 2 ), 'fm arrs must be Nx2 dimensions' with testlog.context('fsv_col_lbls agree with fsv shape'): if cm.fsv_list is None: testlog.skip_test() else: if cm.fsv_col_lbls is not None or strict: assert ( cm.fsv_col_lbls is not None ), 'need to specify the names of the columns' num_col_lbls = len(cm.fsv_col_lbls) else: if len(cm.fsv_list) == 0: num_col_lbls = 0 else: num_col_lbls = cm.fsv_list[0].shape[1] assert ut.list_all_eq_to( [fsv.shape[1] for fsv in cm.fsv_list], num_col_lbls ), 'num_col_lbls=%r' % (num_col_lbls,) with testlog.context('filtnorm checks'): if cm.filtnorm_aids is None and cm.filtnorm_fxs is None: testlog.skip_test() else: with testlog.context('num_col_lbls agree with filtnorm_arrs'): assert ( len(cm.filtnorm_aids) == num_col_lbls ), 'bad len %r != %r' % (len(cm.filtnorm_aids), num_col_lbls) assert len(cm.filtnorm_fxs) == num_col_lbls with testlog.context('len(fsvs) agree with filtnorm_arrs'): assert all( [ aids_list is None or all( [ len(fsv) == len(aids) for aids, fsv in zip(aids_list, cm.fsv_list) ] ) for aids_list in cm.filtnorm_aids ] ), 'norm aid indicies do not agree with featscores' assert all( [ fxs_list is None or all( [ len(fsv) == len(fxs) for fxs, fsv in zip(fxs_list, cm.fsv_list) ] ) for fxs_list in cm.filtnorm_fxs ] ), 'norm fx indicies do not agree with featscores' except Exception: cm.print_rawinfostr() raise # testlog.log_passed('filtkey and fsv shapes are ok') if assert_feats and (strict or qreq_ is not None): external_qaids = aslist(qreq_.qaids) external_daids = aslist(qreq_.daids) proot = getattr(qreq_.qparams, 'pipeline_root', None) if proot == 'vsone': assert len(external_qaids) == 1, 'only one external qaid for vsone' if strict or qreq_.indexer is not None: nExternalQVecs = qreq_.ibs.get_annot_vecs( external_qaids[0], config2_=qreq_.extern_query_config2 ).shape[0] assert ( qreq_.indexer.idx2_vec.shape[0] == nExternalQVecs ), 'did not index query descriptors properly' testlog.log_passed('vsone daids are ok are ok') nFeats1 = qreq_.ibs.get_annot_num_feats( cm.qaid, config2_=qreq_.extern_query_config2 ) nFeats2_list = np.array( qreq_.ibs.get_annot_num_feats( cm.daid_list, config2_=qreq_.extern_data_config2 ) ) if False: # This does not need to be the case especially if the daid_list # was exteneded try: assert ut.list_issubset( cm.daid_list, external_daids ), 'cmtup_old must be subset of daids' except AssertionError as ex: ut.printex(ex, keys=['daid_list', 'external_daids']) raise try: fm_list = cm.fm_list fx2s_list = [fm_.T[1] for fm_ in fm_list] fx1s_list = [fm_.T[0] for fm_ in fm_list] max_fx1_list = np.array( [-1 if len(fx1s) == 0 else fx1s.max() for fx1s in fx1s_list] ) max_fx2_list = np.array( [-1 if len(fx2s) == 0 else fx2s.max() for fx2s in fx2s_list] ) ut.assert_lessthan( max_fx2_list, nFeats2_list, 'max feat index must be less than num feats', ) ut.assert_lessthan( max_fx1_list, nFeats1, 'max feat index must be less than num feats' ) except AssertionError as ex: ut.printex( ex, keys=[ 'qaid', 'daid_list', 'nFeats1', 'nFeats2_list', 'max_fx1_list', 'max_fx2_list', ], ) raise testlog.log_passed('nFeats are ok in fm') else: testlog.log_skipped('nFeat check') if qreq_ is not None: pass
[docs]@ut.reloadable_class class ChipMatch( _ChipMatchVisualization, AnnotMatch, _ChipMatchScorers, old_chip_match._OldStyleChipMatchSimulator, _ChipMatchConvenienceGetter, _ChipMatchDebugger, ): """ behaves as as the ChipMatchOldTup named tuple until we completely replace the old structure """ # Standard Contstructor def __init__(cm, *args, **kwargs): """ qaid and daid_list are not optional. fm_list and fsv_list are strongly encouraged and will probalby break things if they are not there. SeeAlso: initialize """ try: super(ChipMatch, cm).__init__(*args, **kwargs) except TypeError: # Hack for ipython reload logger.info('id(cm.__class__) = %r' % (id(cm.__class__),)) logger.info('id(ChipMatch) = %r' % (id(ChipMatch),)) # import utool # utool.embed() # assert id(cm.__class__) > id(ChipMatch) super(cm.__class__, cm).__init__(*args, **kwargs) if ut.STRICT: raise cm.fm_list = None cm.fsv_list = None cm.fk_list = None cm.fsv_col_lbls = None cm.fs_list = None # Hacks for norm cm.filtnorm_aids = None cm.filtnorm_fxs = None if len(args) + len(kwargs) > 0: cm.initialize(*args, **kwargs)
[docs] def initialize( cm, qaid=None, daid_list=None, fm_list=None, fsv_list=None, fk_list=None, score_list=None, H_list=None, fsv_col_lbls=None, dnid_list=None, qnid=None, unique_nids=None, name_score_list=None, annot_score_list=None, autoinit=True, filtnorm_aids=None, filtnorm_fxs=None, ): """ qaid and daid_list are not optional. fm_list and fsv_list are strongly encouraged and will probalby break things if they are not there. """ if DEBUG_CHIPMATCH: msg = 'incompatable data' assert daid_list is not None, 'must give daids' assert fm_list is None or len(fm_list) == len(daid_list), msg assert fsv_list is None or len(fsv_list) == len(daid_list), msg assert fk_list is None or len(fk_list) == len(daid_list), msg assert H_list is None or len(H_list) == len(daid_list), msg assert score_list is None or len(score_list) == len(daid_list), msg assert dnid_list is None or len(dnid_list) == len(daid_list), msg cm.qaid = qaid cm.daid_list = safeop(np.array, daid_list, dtype=hstypes.INDEX_TYPE) cm.score_list = safeop(np.array, score_list, dtype=hstypes.FLOAT_TYPE) cm.H_list = H_list # name info cm.qnid = qnid cm.dnid_list = safeop(np.array, dnid_list, dtype=hstypes.INDEX_TYPE) cm.unique_nids = safeop(np.array, unique_nids, dtype=hstypes.INDEX_TYPE) cm.name_score_list = safeop(np.array, name_score_list, dtype=hstypes.FLOAT_TYPE) cm.annot_score_list = safeop(np.array, annot_score_list, dtype=hstypes.FLOAT_TYPE) cm.fm_list = fm_list cm.fsv_list = fsv_list cm.fk_list = ( fk_list if fk_list is not None else [np.zeros(fm.shape[0]) for fm in cm.fm_list] if cm.fm_list is not None else None ) cm.fsv_col_lbls = fsv_col_lbls # HACKY normalizer info cm.filtnorm_aids = filtnorm_aids cm.filtnorm_fxs = filtnorm_fxs # TODO: have subclass or dict for special scores if autoinit: cm._update_daid_index() if cm.dnid_list is not None: cm._update_unique_nid_index() if DEBUG_CHIPMATCH: cm.assert_self(verbose=True)
[docs] def arraycast_self(cm): """ Ensures internal structure is in numpy array formats TODO: come up with better name Remove old initialize method and rename to initialize? """ cm.daid_list = safeop(np.array, cm.daid_list, dtype=hstypes.INDEX_TYPE) cm.score_list = safeop(np.array, cm.score_list, dtype=hstypes.FLOAT_TYPE) # name info cm.dnid_list = safeop(np.array, cm.dnid_list, dtype=hstypes.INDEX_TYPE) cm.unique_nids = safeop(np.array, cm.unique_nids, dtype=hstypes.INDEX_TYPE) cm.name_score_list = safeop( np.array, cm.name_score_list, dtype=hstypes.FLOAT_TYPE ) cm.annot_score_list = safeop( np.array, cm.annot_score_list, dtype=hstypes.FLOAT_TYPE ) ncols = None if cm.fsv_col_lbls is None else len(cm.fsv_col_lbls) cm.H_list = safecast_numpy_lists(cm.H_list, dtype=hstypes.FLOAT_TYPE) cm.fm_list = safecast_numpy_lists( cm.fm_list, dtype=hstypes.INDEX_TYPE, dims=(None, 2) ) cm.fsv_list = safecast_numpy_lists( cm.fsv_list, dtype=hstypes.FLOAT_TYPE, dims=(None, ncols) ) cm.fk_list = safecast_numpy_lists(cm.fk_list, dtype=hstypes.INDEX_TYPE)
def _empty_hack(cm): if cm.daid_list is None: cm.daid_list = np.empty(0, dtype=np.int) assert len(cm.daid_list) == 0 cm.fsv_col_lbls = [] cm.fm_list = [] cm.fsv_list = [] cm.fk_list = [] cm.H_list = [] cm.daid2_idx = {} cm.fs_list = [] cm.dnid_list = np.empty(0, dtype=hstypes.INDEX_TYPE) cm.unique_nids = np.empty(0, dtype=hstypes.INDEX_TYPE) cm.score_list = np.empty(0) cm.name_score_list = np.empty(0) cm.annot_score_list = np.empty(0) def __eq__(cm, other): # if isinstance(other, cm.__class__): try: flag = True flag &= len(cm.fm_list) == len(other.fm_list) flag &= cm.qaid == other.qaid flag &= cm.qnid == other.qnid flag &= check_arrs_eq(cm.fm_list, other.fm_list) flag &= check_arrs_eq(cm.fs_list, other.fs_list) flag &= check_arrs_eq(cm.fk_list, other.fk_list) flag &= check_arrs_eq(cm.daid_list, other.daid_list) flag &= check_arrs_eq(cm.dnid_list, other.dnid_list) flag &= check_arrs_eq(cm.unique_nids, other.unique_nids) return flag except AttributeError: return False # else: # return False # ------------------ # Modification / Evaluation Functions # ------------------ def _cast_scores(cm, dtype=np.float64): cm.fsv_list = [fsv.astype(dtype) for fsv in cm.fsv_list]
[docs] def compress_results(cm, inplace=False): flags = [len(fm) > 1 for fm in cm.fm_list] out = cm.compress_annots(flags, inplace=inplace) return out
[docs] def extend_results(cm, qreq_, other_aids=None): """ Return a new ChipMatch containing empty data for an extended set of aids Args: qreq_ (wbia.QueryRequest): query request object with hyper-parameters other_aids (None): (default = None) Returns: wbia.ChipMatch: out CommandLine: python -m wbia.algo.hots.chip_match --exec-extend_results --show Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', >>> a='default:dindex=0:10,qindex=0:1', >>> t='best:SV=False') >>> assert len(cm.daid_list) == 9 >>> cm.assert_self(qreq_) >>> other_aids = qreq_.ibs.get_valid_aids() >>> out = cm.extend_results(qreq_, other_aids) >>> assert len(out.daid_list) == 118 >>> out.assert_self(qreq_) """ if other_aids is None: other_aids = qreq_.daids ibs = qreq_.ibs other_aids_ = other_aids other_aids_ = np.setdiff1d(other_aids_, cm.daid_list) other_aids_ = np.setdiff1d(other_aids_, [cm.qaid]) other_nids_ = ibs.get_annot_nids(other_aids_) other_unique_nids = np.setdiff1d(np.unique(other_nids_), cm.unique_nids) num = len(other_aids_) num2 = len(other_unique_nids) daid_list = np.append(cm.daid_list, other_aids_) dnid_list = np.append(cm.dnid_list, other_nids_) score_list = extend_scores(cm.score_list, num) annot_score_list = extend_scores(cm.annot_score_list, num) unique_nids = np.append(cm.unique_nids, other_unique_nids) name_score_list = extend_scores(cm.name_score_list, num2) qaid = cm.qaid qnid = cm.qnid fsv_col_lbls = cm.fsv_col_lbls # <feat correspondence> nVs = 0 if fsv_col_lbls is None else len(fsv_col_lbls) fm_list = extend_nplists(cm.fm_list, num, (0, 2), hstypes.FM_DTYPE) fk_list = extend_nplists(cm.fk_list, num, (0), hstypes.FK_DTYPE) fs_list = extend_nplists(cm.fs_list, num, (0), hstypes.FS_DTYPE) fsv_list = extend_nplists(cm.fsv_list, num, (0, nVs), hstypes.FS_DTYPE) H_list = extend_pylist(cm.H_list, num, None) filtnorm_aids = filtnorm_op( cm.filtnorm_aids, extend_nplists, num, (0), hstypes.INDEX_TYPE ) filtnorm_fxs = filtnorm_op( cm.filtnorm_fxs, extend_nplists, num, (0), hstypes.INDEX_TYPE ) # </feat correspondence> out = ChipMatch( qaid, daid_list, fm_list, fsv_list, fk_list, score_list, H_list, fsv_col_lbls, dnid_list, qnid, unique_nids, name_score_list, annot_score_list, filtnorm_fxs=filtnorm_fxs, filtnorm_aids=filtnorm_aids, autoinit=False, ) out.fs_list = fs_list # attrs should be dicts for key in cm.algo_annot_scores.keys(): out.algo_annot_scores[key] = extend_scores(cm.algo_annot_scores[key], num) for key in cm.algo_name_scores.keys(): out.algo_name_scores[key] = extend_scores(cm.algo_name_scores[key], num2) out._update_daid_index() out._update_unique_nid_index() return out
[docs] @classmethod def combine_cms(ChipMatch, cm_list): """ Example: >>> # FIXME failing-test (22-Jul-2020) This test is failing and it's not clear how to fix it >>> # xdoctest: +SKIP >>> from wbia.core_annots import * # NOQA >>> ibs, depc, aid_list = testdata_core(size=4) >>> request = depc.new_request('vsone', [1], [2, 3, 4], {'dim_size': 450}) >>> rawres_list2 = request.execute(postprocess=False) >>> cm_list = ut.take_column(rawres_list2, 1) >>> out = ChipMatch.combine_cms(cm_list) >>> out.score_name_nsum(request) >>> ut.quit_if_noshow() >>> out.ishow_analysis(request) >>> ut.show_if_requested() """ new_attrs = {} common_attrs = ['qaid', 'qnid', 'fsv_col_lbls'] for attr in common_attrs: values = ut.list_getattr(cm_list, 'qaid') assert ut.allsame(values) new_attrs[attr] = values[0] # assumes disjoint attrs = [ 'daid_list', 'dnid_list', 'score_list', 'annot_score_list', 'H_list', 'fm_list', 'fsv_list', 'fk_list', 'filtnorm_aids', 'filtnorm_fxs', ] new_attrs['qaid'] = cm_list[0].qaid new_attrs['qnid'] = cm_list[0].qnid new_attrs['fsv_col_lbls'] = cm_list[0].fsv_col_lbls for attr in attrs: values = ut.list_getattr(cm_list, attr) if ut.list_all_eq_to(values, None): new_attrs[attr] = None else: new_attrs[attr] = ut.flatten(values) out = ChipMatch(**new_attrs) out._update_daid_index() out._update_unique_nid_index() return out
[docs] def take_annots(cm, idx_list, inplace=False, keepscores=True): """ Keeps results only for the selected annotation indices. CommandLine: python -m wbia.algo.hots.chip_match take_annots Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', >>> a='default:dindex=0:10,qindex=0:1', >>> t='best:sv=False') >>> idx_list = list(range(cm.num_daids)) >>> inplace = False >>> keepscores = True >>> other = out = cm.take_annots(idx_list, inplace, keepscores) >>> result = ('out = %s' % (ut.repr2(out, nl=1),)) >>> # Because the subset was all aids in order, the output >>> # ChipMatch should be exactly the same. >>> assert cm.inspect_difference(out), 'Should be exactly equal!' >>> print(result) Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', >>> a='default:dindex=0:10,qindex=0:1', >>> t='best:SV=False') >>> idx_list = [0, 2] >>> inplace = False >>> keepscores = True >>> other = out = cm.take_annots(idx_list, inplace, keepscores) >>> result = ('out = %s' % (ut.repr2(out, nl=1),)) >>> print(result) """ if inplace: out = cm else: out = ChipMatch(qaid=cm.qaid, qnid=cm.qnid, fsv_col_lbls=cm.fsv_col_lbls) out.daid_list = vt.take2(cm.daid_list, idx_list) out.dnid_list = safeop(vt.take2, cm.dnid_list, idx_list) out.H_list = safeop(ut.take, cm.H_list, idx_list) out.fm_list = safeop(ut.take, cm.fm_list, idx_list) out.fsv_list = safeop(ut.take, cm.fsv_list, idx_list) out.fk_list = safeop(ut.take, cm.fk_list, idx_list) out.filtnorm_aids = filtnorm_op(cm.filtnorm_aids, ut.take, idx_list) out.filtnorm_fxs = filtnorm_op(cm.filtnorm_fxs, ut.take, idx_list) if keepscores: # Annot Scores out.score_list = safeop(vt.take2, cm.score_list, idx_list) out.annot_score_list = safeop(vt.take2, cm.annot_score_list, idx_list) for key in out.algo_annot_scores.keys(): out.algo_annot_scores[key] = safeop( vt.take2, cm.algo_annot_scores[key], idx_list ) # Name Scores if True: nidxs_subset = ut.take(cm.nid2_nidx, np.unique(out.dnid_list)) out.unique_nids = safeop(vt.take2, cm.unique_nids, nidxs_subset) out.name_score_list = safeop(vt.take2, cm.name_score_list, nidxs_subset) for key in out.algo_name_scores.keys(): subset = safeop(vt.take2, cm.algo_name_scores[key], nidxs_subset) out.algo_name_scores[key] = subset out.nid2_nidx = None out.name_groupxs = None else: # Name Scores # TODO: remove score of names that were removed? out.nid2_nidx = cm.nid2_nidx out.unique_nids = cm.unique_nids out.name_score_list = cm.name_score_list for key in out.algo_name_scores.keys(): out.algo_name_scores[key] = cm.algo_name_scores[key] out._update_daid_index() out._update_unique_nid_index() return out
[docs] def take_feature_matches(cm, indicies_list, inplace=False, keepscores=True): r""" Removes outlier feature matches TODO: rectify with shortlist_subset Args: indicies_list (list): list of lists of indicies to keep. if an item is None, the match to the corresponding daid is removed. inplace (bool): (default = False) Returns: wbia.ChipMatch: out CommandLine: python -m wbia.algo.hots.chip_match --exec-take_feature_matches --show Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm('PZ_MTEST', a='default:dindex=0:10,qindex=0:1', t='best:SV=False') >>> indicies_list = [list(range(i + 1)) for i in range(cm.num_daids)] >>> inplace = False >>> keepscores = True >>> out = cm.take_feature_matches(indicies_list, inplace, keepscores) >>> assert not cm.inspect_difference(out, verbose=False), 'should be different' >>> result = ('out = %s' % (ut.repr2(out),)) >>> print(result) """ assert len(indicies_list) == len(cm.daid_list), 'must correspond to daids' flags = ut.flag_not_None_items(indicies_list) # Remove disgarded matches out = cm.compress_annots(flags, inplace=inplace, keepscores=keepscores) indicies_list2 = ut.compress(indicies_list, flags) out.fm_list = safeop(vt.ziptake, out.fm_list, indicies_list2, axis=0) out.fs_list = safeop(vt.ziptake, out.fs_list, indicies_list2, axis=0) out.fsv_list = safeop(vt.ziptake, out.fsv_list, indicies_list2, axis=0) out.fk_list = safeop(vt.ziptake, out.fk_list, indicies_list2, axis=0) out.filtnorm_aids = filtnorm_op( out.filtnorm_aids, vt.ziptake, indicies_list2, axis=0 ) out.filtnorm_fxs = filtnorm_op( out.filtnorm_fxs, vt.ziptake, indicies_list2, axis=0 ) # out.assert_self(verbose=False) return out
[docs] def shortlist_subset(cm, top_aids): """returns a new cmtup_old with only the requested daids TODO: rectify with take_feature_matches """ idx_list = ut.dict_take(cm.daid2_idx, top_aids) out = cm.take_annots(idx_list, keepscores=False) return out
[docs] def compress_annots(cm, flags, inplace=False, keepscores=True): idx_list = np.where(flags)[0] out = cm.take_annots(idx_list, inplace, keepscores) return out
[docs] def append_featscore_column(cm, filtkey, filtweight_list, inplace=True): assert inplace, 'this is always inplace right now' assert filtkey not in cm.fsv_col_lbls, 'already have filtkey=%r' % (cm.filtkey,) cm.fsv_col_lbls.append(filtkey) cm.fsv_list = vt.zipcat(cm.fsv_list, filtweight_list, axis=1)
[docs] def compress_top_feature_matches(cm, num=10, rng=np.random, use_random=True): """ DO NOT USE FIXME: Use boolean lists Removes all but the best feature matches for testing purposes rng = np.random.RandomState(0) """ # num = 10 fs_list = cm.get_fsv_prod_list() score_sortx = [fs.argsort()[::-1] for fs in fs_list] if use_random: # keep jagedness score_sortx_filt = [ sortx[0 : min(rng.randint(num // 2, num), len(sortx))] for sortx in score_sortx ] else: score_sortx_filt = [sortx[0 : min(num, len(sortx))] for sortx in score_sortx] # cm.take_feature_matches() cm.fsv_list = vt.ziptake(cm.fsv_list, score_sortx_filt, axis=0) cm.fm_list = vt.ziptake(cm.fm_list, score_sortx_filt, axis=0) cm.fk_list = vt.ziptake(cm.fk_list, score_sortx_filt, axis=0) if cm.fs_list is not None: cm.fs_list = vt.ziptake(cm.fs_list, score_sortx_filt, axis=0) cm.H_list = None cm.fs_list = None
[docs] def sortself(cm): """reorders the internal data using cm.score_list""" logger.info('Warning using sortself') sortx = cm.argsort() cm.daid_list = vt.trytake(cm.daid_list, sortx) cm.dnid_list = vt.trytake(cm.dnid_list, sortx) cm.fm_list = vt.trytake(cm.fm_list, sortx) cm.fsv_list = vt.trytake(cm.fsv_list, sortx) cm.fs_list = vt.trytake(cm.fs_list, sortx) cm.fk_list = vt.trytake(cm.fk_list, sortx) cm.score_list = vt.trytake(cm.score_list, sortx) # FIXME: Not all properties covered cm.algo_annot_scores['csum'] = vt.trytake(cm.algo_annot_scores['csum'], sortx) cm.H_list = vt.trytake(cm.H_list, sortx) cm._update_daid_index()
# --- # Alternative Cosntructors / Convertors
[docs] @classmethod def from_json(ChipMatch, json_str): r""" Convert json string back to ChipMatch object CommandLine: # FIXME: util_test is broken with classmethods python -m wbia.algo.hots.chip_match --test-from_json --show Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm1, qreq_ = wbia.testdata_cm() >>> json_str = cm1.to_json() >>> cm = ChipMatch.from_json(json_str) >>> ut.quit_if_noshow() >>> cm.score_name_nsum(qreq_) >>> cm.show_single_namematch(qreq_, 1) >>> ut.show_if_requested() """ class_dict = ut.from_json(json_str) return ChipMatch.from_dict(class_dict)
[docs] @classmethod def from_dict(ChipMatch, class_dict, ibs=None): r""" Convert dict of arguments back to ChipMatch object """ key_list = ut.get_kwargs(ChipMatch.initialize)[0] # HACKY key_list.remove('autoinit') if ut.VERBOSE: other_keys = list(set(class_dict.keys()) - set(key_list)) if len(other_keys) > 0: logger.info( 'Not unserializing extra attributes: %s' % (ut.repr2(other_keys)) ) if ibs is not None: class_dict = prepare_dict_uuids(class_dict, ibs) dict_subset = ut.dict_subset(class_dict, key_list) dict_subset['fm_list'] = convert_numpy_lists( dict_subset['fm_list'], hstypes.FM_DTYPE, dims=2 ) dict_subset['fsv_list'] = convert_numpy_lists( dict_subset['fsv_list'], hstypes.FS_DTYPE, dims=2 ) dict_subset['score_list'] = convert_numpy( dict_subset['score_list'], hstypes.FS_DTYPE ) safe_check_nested_lens_eq(dict_subset['fm_list'], dict_subset['fsv_list']) safe_check_lens_eq(dict_subset['score_list'], dict_subset['fsv_list']) safe_check_lens_eq(dict_subset['score_list'], dict_subset['fm_list']) cm = ChipMatch(**dict_subset) return cm
[docs] @profile def to_json(cm): r""" Serialize ChipMatch object as JSON string CommandLine: python -m wbia.algo.hots.chip_match --test-ChipMatch.to_json:0 python -m wbia.algo.hots.chip_match --test-ChipMatch.to_json python -m wbia.algo.hots.chip_match --test-ChipMatch.to_json:1 --show Example: >>> # ENABLE_DOCTEST >>> # Simple doctest demonstrating the json format >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm() >>> cm.compress_top_feature_matches(num=4, rng=np.random.RandomState(0)) >>> # Serialize >>> print('\n\nRaw ChipMatch JSON:\n') >>> json_str = cm.to_json() >>> print(json_str) >>> print('\n\nPretty ChipMatch JSON:\n') >>> # Pretty String Formatting >>> dictrep = ut.from_json(json_str) >>> dictrep = ut.delete_dict_keys(dictrep, [key for key, val in dictrep.items() if val is None]) >>> result = ut.repr2_json(dictrep, nl=2, precision=2, key_order_metric='strlen') >>> print(result) Example: >>> # ENABLE_DOCTEST >>> # test to convert back and forth from json >>> from wbia.algo.hots.chip_match import * # NOQA >>> import wbia >>> cm, qreq_ = wbia.testdata_cm() >>> cm1 = cm >>> # Serialize >>> json_str = cm.to_json() >>> print(repr(json_str)) >>> # Unserialize >>> cm = ChipMatch.from_json(json_str) >>> # Show if it works >>> ut.quit_if_noshow() >>> cm.score_name_nsum(qreq_) >>> cm.show_single_namematch(qreq_, 1) >>> ut.show_if_requested() >>> # result = ('json_str = \n%s' % (str(json_str),)) >>> # print(result) """ data = cm.__dict__.copy() # can't encode dictionaries with integer keys # this means you need to rebuild indexes on reconstruction ut.delete_dict_keys(data, ['daid2_idx', 'nid2_nidx']) # logger.info('data = %r' % (list(data.keys()),)) json_str = ut.to_json(data) return json_str
# --- IO
[docs] def get_fpath(cm, qreq_): dpath = qreq_.get_qresdir() fname = get_chipmatch_fname(cm.qaid, qreq_) fpath = join(dpath, fname) return fpath
[docs] def save(cm, qreq_, verbose=None): fpath = cm.get_fpath(qreq_) cm.save_to_fpath(fpath, verbose=verbose)
# @classmethod # def load(cls, qreq_, qaid, dpath=None, verbose=None): # fname = get_chipmatch_fname(qaid, qreq_) # if dpath is None: # dpath = qreq_.get_qresdir() # fpath = join(dpath, fname) # cm = cls.load_from_fpath(fpath, verbose=verbose) # return cm
[docs] @classmethod def load_from_fpath(ChipMatch, fpath, verbose=None): # state_dict = ut.load_data(fpath, verbose=verbose) state_dict = ut.load_cPkl(fpath, verbose=verbose) if 'filtnorm_aids' not in state_dict: raise NeedRecomputeError('old version of chipmatch') cm = ChipMatch() cm.__setstate__(state_dict) return cm
# ----- # Misc # -----
[docs]class TestLogger(object): def __init__(testlog, verbose=True): testlog.test_out = ut.ddict(list) testlog.current_test = None testlog.failed_list = [] testlog.verbose = verbose
[docs] def start_test(testlog, name): testlog.current_test = name
[docs] def log_skipped(testlog, msg): if testlog.verbose: logger.info('[cm] skip: ' + msg)
[docs] def log_passed(testlog, msg): if testlog.verbose: logger.info('[cm] pass: ' + msg)
[docs] def skip_test(testlog): testlog.log_skipped(testlog.current_test) testlog.current_test = None
[docs] def log_failed(testlog, msg): testlog.test_out[testlog.current_test].append(msg) testlog.failed_list.append(msg) logger.info('[cm] FAILED!: ' + msg)
[docs] def end_test(testlog): if len(testlog.test_out[testlog.current_test]) == 0: testlog.log_passed(testlog.current_test) else: testlog.log_failed(testlog.current_test) testlog.current_test = None
[docs] def context(testlog, name): testlog.start_test(name) return testlog
def __enter__(testlog): return testlog def __exit__(testlog, type_, value, trace): if testlog.current_test is not None: if trace is not None: testlog.log_failed('error occured') testlog.end_test()
[docs]def testdata_cm(): ibs, qreq_, cm_list = plh.testdata_pre_sver('PZ_MTEST', qaid_list=[18]) cm = cm_list[0] cm.score_name_nsum(qreq_) return cm, qreq_
[docs]@profile def get_chipmatch_fname( qaid, qreq_, qauuid=None, cfgstr=None, TRUNCATE_UUIDS=TRUNCATE_UUIDS, MAX_FNAME_LEN=MAX_FNAME_LEN, ): r""" CommandLine: python -m wbia.algo.hots.chip_match --test-get_chipmatch_fname Example: >>> # ENABLE_DOCTEST >>> from wbia.algo.hots.chip_match import * # NOQA >>> qreq_, args = plh.testdata_pre('spatial_verification', >>> defaultdb='PZ_MTEST', qaid_override=[18], >>> p='default:sqrd_dist_on=True') >>> cm_list = args.cm_list_FILT >>> cm = cm_list[0] >>> fname = get_chipmatch_fname(cm.qaid, qreq_, qauuid=None, >>> TRUNCATE_UUIDS=False, MAX_FNAME_LEN=200) >>> result = fname >>> print(result) qaid=18_cm_cvgrsbnffsgifyom_quuid=a126d459-b730-573e-7a21-92894b016565.cPkl """ if qauuid is None: logger.info('[chipmatch] Warning: qasuuid should be given') qauuid = next(qreq_.get_qreq_pcc_uuids([qaid])) if cfgstr is None: logger.info('[chipmatch] Warning: cfgstr should be passed given') cfgstr = qreq_.get_cfgstr(with_input=True) # logger.info('cfgstr = %r' % (cfgstr,)) fname_fmt = 'qaid={qaid}_cm_{cfgstr}_quuid={qauuid}{ext}' qauuid_str = str(qauuid)[0:8] if TRUNCATE_UUIDS else str(qauuid) fmt_dict = dict(cfgstr=cfgstr, qaid=qaid, qauuid=qauuid_str, ext='.cPkl') fname = ut.long_fname_format( fname_fmt, fmt_dict, ['cfgstr'], max_len=MAX_FNAME_LEN, hack27=True ) return fname