def get_dataset(cf):
dataset = Dataset_daly_ocv(cf['dataset.mirror'])
dataset.populate_from_folder(cf['dataset.cache_folder'])
# Augment video with split
for vid, split_name in dataset.split.items():
dataset.videos_ocv[vid]['split'] = split_name
return dataset
def eval_daly_object(workfolder, cfg_dict, add_args):
"""
Evaluation code with hacks
"""
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
_set_defcfg_detectron(cfg)
_set_defcfg_detectron_test(cfg)
_set_defcfg_object_hacks(cfg)
cfg.set_deftype("""
what_to_eval: [~, str]
eval_hacks:
model_to_eval: ['what', ['what', 'what+foldname']]
""")
cf = cfg.parse()
cf_add_d2 = cfg.without_prefix('d2.')
# DALY Dataset
dataset = Dataset_daly_ocv()
dataset.populate_from_folder(cf['dataset.cache_folder'])
split_label = cf['dataset.subset']
split_vids = get_daly_split_vids(dataset, split_label)
datalist: Datalist = simplest_daly_to_datalist_v2(dataset, split_vids)
cls_names, datalist_converter = \
_datalist_hacky_converter(cf, dataset)
datalist = datalist_converter(datalist)
TEST_DATASET_NAME = 'daly_objaction_test'
model_to_eval = _eval_foldname_hack(cf)
_eval_routine(cf, cf_add_d2, out, cls_names,
TEST_DATASET_NAME, datalist, model_to_eval)
def train_daly_object(workfolder, cfg_dict, add_args):
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
_set_defcfg_detectron(cfg)
_set_defcfg_detectron_train(cfg)
_set_defcfg_object_hacks(cfg)
cf = cfg.parse()
cf_add_d2 = cfg.without_prefix('d2.')
cf_add_d2 = d2dict_gpu_scaling(cf, cf_add_d2, cf['num_gpus'])
dataset = Dataset_daly_ocv()
dataset.populate_from_folder(cf['dataset.cache_folder'])
o100_objects, category_map = get_category_map_o100(dataset)
assert len(o100_objects) == 16
split_label = cf['dataset.subset']
split_vids = get_daly_split_vids(dataset, split_label)
datalist: Datalist = simplest_daly_to_datalist_v2(dataset, split_vids)
cls_names, datalist_converter = \
_datalist_hacky_converter(cf, dataset)
datalist = datalist_converter(datalist)
TRAIN_DATASET_NAME = 'daly_objaction_train'
_train_routine(cf, cf_add_d2, out,
cls_names, TRAIN_DATASET_NAME, datalist, add_args)
def compare_data(workfolder, cfg_dict, add_args):
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
cfg.set_defaults_handling(raise_without_defaults=False)
cfg.set_defaults("""
cache_folders: ~
""")
cf = cfg.parse()
dataset = Dataset_daly_ocv()
dataset.precompute_to_folder(out)
def resolve_dataset_tubes(cf):
dataset = Dataset_daly_ocv(cf['dataset.mirror'])
dataset.populate_from_folder(cf['dataset.cache_folder'])
split_label = cf['dataset.subset']
split_vids: List[Vid_daly] = \
get_daly_split_vids(dataset, split_label)
dgt_tubes: Dict[I_dgt, T_dgt] = \
get_daly_gt_tubes(dataset)
dgt_tubes = dtindex_filter_split(dgt_tubes, split_vids)
av_gt_tubes: AV_dict[T_dgt] = push_into_avdict(dgt_tubes)
return dataset, split_vids, av_gt_tubes
def load_wein_tubes(workfolder, cfg_dict, add_args):
"""
Philippe tubes:
tube:
(one row per frame):
index of the frame (starting at 1)
x1 y1 x2 y2
score of the generic human detector
score of the instance-specific detector
"""
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
cfg.set_deftype("""
wein_tubes: [~, str]
dataset:
cache_folder: [~, str]
mirror: ['uname', ~]
wein_index: [0, int] # 0 = 510 cut tubes, 1 = 200 uncut tubes
""")
cf = cfg.parse()
# Dataset
dataset = Dataset_daly_ocv(cf['dataset.mirror'])
dataset.populate_from_folder(cf['dataset.cache_folder'])
# Tubes
# 0 has 510 elements, 1 has 200 elements
wein_package = small.load_py2_pkl(cf['wein_tubes'])[cf['wein_index']]
# We got a dictionary of filenames (w .mp4 suffix)
extracted_tubes: Dict[I_dwein, Tube_daly_wein_as_provided] = {}
for vid_mp4, wein_bunches in wein_package.items():
vid = re.search(r'(.*)\.mp4', vid_mp4).group(1)
rs = dataset.rstats[vid]
for bunch_id, wein_tubes in enumerate(wein_bunches):
for tube_id, wein_tube in enumerate(wein_tubes):
frame_inds = wein_tube[:, 0].astype(np.int) - 1
assert max(frame_inds) < rs['max_pos_frames']
boxes_ltrd = wein_tube[:, 1:5] # ltrd
human_scores = wein_tube[:, 5]
instance_scores = wein_tube[:, 6]
tube = {
'frame_inds': frame_inds,
'boxes': boxes_ltrd,
'hscores': human_scores,
'iscores': instance_scores
}
extracted_tubes[(vid, bunch_id, tube_id)] = tube
small.save_pkl(out / 'extracted_tubes.pkl', extracted_tubes)
def daly_oldstats(workfolder, cfg_dict, add_args):
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.yconfig.YConfig_v2(cfg_dict, allowed_wo_defaults=[''])
cfg.set_defaults_yaml("""
dataset:
cache_folder: ~
mirror: 'uname'
""")
cf = cfg.parse()
dataset = Dataset_daly_ocv(cf['dataset.mirror'])
dataset.populate_from_folder(cf['dataset.cache_folder'])
pd_kf, pd_obj = Daly_stats.pd_stats(dataset)
Daly_stats.inst_bbox_stats_hists(pd_kf, out)
Daly_stats.inst_bbox_stats_hmap(pd_kf, out, dataset)
Daly_stats.obj_inst_bbox_stats(pd_obj, out, dataset)
def train_daly_action(workfolder, cfg_dict, add_args):
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
_set_defcfg_detectron(cfg)
_set_defcfg_detectron_train(cfg)
cf = cfg.parse()
cf_add_d2 = cfg.without_prefix('d2.')
cf_add_d2 = d2dict_gpu_scaling(cf, cf_add_d2, cf['num_gpus'])
dataset = Dataset_daly_ocv()
dataset.populate_from_folder(cf['dataset.cache_folder'])
split_label = cf['dataset.subset']
split_vids = get_daly_split_vids(dataset, split_label)
datalist: Datalist = daly_to_datalist_pfadet(dataset, split_vids)
cls_names = dataset.action_names
TRAIN_DATASET_NAME = 'daly_pfadet_train'
_train_routine(cf, cf_add_d2, out,
cls_names, TRAIN_DATASET_NAME, datalist, add_args)
def precompute_cache(workfolder, cfg_dict, add_args):
out, = snippets.get_subfolders(workfolder, ['out'])
cfg = snippets.YConfig(cfg_dict)
cfg.set_deftype("""
dataset: [~, str]
charades:
mirror: ['gpuhost7', str]
resolution: ['480', str]
daly:
mirror: [~, str]
""")
cf = cfg.parse()
if cf['dataset'] == 'daly':
dataset = Dataset_daly_ocv(cf['daly.mirror'])
elif cf['dataset'] == 'charades':
dataset = Dataset_charades_ocv(cf['charades.mirror'],
cf['charades.resolution'])
elif cf['dataset'] == 'voc2007':
raise NotImplementedError()
else:
raise RuntimeError('Wrong dataset')
dataset.precompute_to_folder(out)