-
Notifications
You must be signed in to change notification settings - Fork 3
/
Copy pathevaluate.py
181 lines (162 loc) · 7.82 KB
/
evaluate.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
import os
import json
import argparse
from collections import defaultdict
from tqdm import tqdm
from dataset import VidVRD, VidOR
from evaluation import eval_video_object, eval_action, eval_visual_relation, print_relation_scores
from visualize import visualize
def evaluate_object(dataset, split, prediction):
groundtruth = dict()
for vid in dataset.get_index(split):
groundtruth[vid] = dataset.get_object_insts(vid)
mean_ap, ap_class = eval_video_object(groundtruth, prediction)
def evaluate_action(dataset, split, prediction):
groundtruth = dict()
for vid in dataset.get_index(split):
groundtruth[vid] = dataset.get_action_insts(vid)
mean_ap, ap_class = eval_action(groundtruth, prediction)
def evaluate_relation(dataset, split, prediction):
scores = dict()
print('[info] evaluating overall setting')
groundtruth = dict()
for vid in dataset.get_index(split):
groundtruth[vid] = dataset.get_relation_insts(vid)
scores['overall'] = eval_visual_relation(groundtruth, prediction)
for use_origin_zeroshot_eval in [False, True]:
if use_origin_zeroshot_eval:
print('[info] evaluating generalized zero-shot setting')
else:
print('[info] evaluating zero-shot setting')
zeroshot_triplets = dataset.get_triplets(split).difference(
dataset.get_triplets('train'))
groundtruth = dict()
zs_prediction = dict()
for vid in dataset.get_index(split):
gt_relations = dataset.get_relation_insts(vid)
zs_gt_relations = []
for r in gt_relations:
if tuple(r['triplet']) in zeroshot_triplets:
zs_gt_relations.append(r)
if len(zs_gt_relations) > 0:
groundtruth[vid] = zs_gt_relations
if use_origin_zeroshot_eval:
# origin zero-shot evaluation doesn't filter out non-zeroshot predictions
# in a video, which is the generalized zero-shot setting
zs_prediction[vid] = prediction[vid]
else:
zs_prediction[vid] = []
for r in prediction.get(vid, []):
if tuple(r['triplet']) in zeroshot_triplets:
zs_prediction[vid].append(r)
if use_origin_zeroshot_eval:
scores['generalized zero-shot'] = eval_visual_relation(groundtruth, zs_prediction)
else:
scores['zero-shot'] = eval_visual_relation(groundtruth, zs_prediction)
return scores
def convert_format(anno, pred_relations, pred_version):
normalize_coords = pred_version >= 'VERSION 2.1'
entities = []
trajectories = defaultdict(list)
relation_instances = []
if normalize_coords:
width_ratio = anno['width']
height_ratio = anno['height']
else:
width_ratio = height_ratio = 1
if pred_version >= 'VERSION 3.0':
for fid in range(len(pred_relations['trajectories'])):
frame = pred_relations['trajectories'][fid]
for region in frame:
region['bbox']['xmin'] *= width_ratio
region['bbox']['ymin'] *= height_ratio
region['bbox']['xmax'] *= width_ratio
region['bbox']['ymax'] *= height_ratio
trajectories[fid] = frame
entities = pred_relations['subject/objects']
relation_instances = pred_relations['relation_instances']
else:
for rel_inst in pred_relations:
tid = {}
tid['sub_traj'] = len(entities)
entities.append({
'tid': tid['sub_traj'],
'category': rel_inst['triplet'][0]
})
tid['obj_traj'] = len(entities)
entities.append({
'tid': tid['obj_traj'],
'category': rel_inst['triplet'][2]
})
fstart, fend = rel_inst['duration']
relation_instances.append({
'subject_tid': tid['sub_traj'],
'object_tid': tid['obj_traj'],
'predicate': rel_inst['triplet'][1],
'score': rel_inst['score'],
'begin_fid': fstart,
'end_fid': fend
})
for e in ['sub_traj', 'obj_traj']:
for i, bbox in enumerate(rel_inst[e]):
trajectories[fstart+i].append({
'tid': tid[e],
'bbox': {
'xmin': bbox[0]*width_ratio,
'ymin': bbox[1]*height_ratio,
'xmax': bbox[2]*width_ratio,
'ymax': bbox[3]*height_ratio
}
})
anno = dict(anno)
anno['subject/objects'] = entities
anno['trajectories'] = [trajectories[fid] for fid in range(anno['frame_count'])]
anno['relation_instances'] = relation_instances
return anno
if __name__ == '__main__':
parser = argparse.ArgumentParser(description='Evaluate a set of tasks related to video relation understanding.')
parser.add_argument('dataset', type=str, help='the dataset name for evaluation')
parser.add_argument('split', type=str, help='the split name for evaluation')
parser.add_argument('task', choices=['object', 'action', 'relation'], help='which task to evaluate')
parser.add_argument('prediction', type=str, help='Corresponding prediction JSON file')
parser.add_argument('--visualize', action="store_true", default=False, help='Visualize for qualitative evaluation')
args = parser.parse_args()
print('[info] loading prediction from {}'.format(args.prediction))
with open(args.prediction, 'r') as fin:
pred = json.load(fin)
print('------ number of videos in prediction: {}'.format(len(pred['results'])))
normalize_coords = pred['version'] >= 'VERSION 2.1'
if args.dataset=='imagenet-vidvrd':
if args.task=='relation':
# load train set for zero-shot evaluation
dataset = VidVRD('../imagenet-vidvrd-dataset', '../imagenet-vidvrd-dataset/videos', ['train', args.split], normalize_coords=normalize_coords)
else:
dataset = VidVRD('../imagenet-vidvrd-dataset', '../imagenet-vidvrd-dataset/videos', [args.split], normalize_coords=normalize_coords)
elif args.dataset=='vidor':
if args.task=='relation':
# load train set for zero-shot evaluation
dataset = VidOR('../vidor-dataset/annotation', '../vidor-dataset/video', ['training', args.split], low_memory=True, normalize_coords=normalize_coords)
else:
dataset = VidOR('../vidor-dataset/annotation', '../vidor-dataset/video', [args.split], low_memory=True, normalize_coords=normalize_coords)
else:
raise Exception('Unknown dataset {}'.format(args.dataset))
if args.task=='object':
assert args.vis_path is None, 'not implemented'
evaluate_object(dataset, args.split, pred['results'])
elif args.task=='action':
assert args.vis_path is None, 'not implemented'
evaluate_action(dataset, args.split, pred['results'])
elif args.task=='relation':
if args.visualize:
vis_path = os.path.join(os.path.dirname(args.prediction), 'visualize')
if not os.path.exists(vis_path):
os.mkdir(vis_path)
for vid, pred_relations in tqdm(pred['results'].items()):
anno = dataset.get_anno(vid)
vis_anno = convert_format(anno, pred_relations, pred['version'])
video_path = dataset.get_video_path(vid)
out_path = os.path.join(vis_path, '{}.mp4'.format(vid))
visualize(vis_anno, video_path, out_path, relation_panel=True, bbox_perturb=0.02)
else:
scores = evaluate_relation(dataset, args.split, pred['results'])
print_relation_scores(scores)