-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathinference.py
239 lines (207 loc) · 11.9 KB
/
inference.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
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
# ------------------------------------------------------------------------
# InstanceFormer Inference.
# ------------------------------------------------------------------------
import argparse
import random
import torch
import glob
import util.misc as utils
from models import build_model
import torchvision.transforms as T
import os
from PIL import Image
import torch.nn.functional as F
import json
import pycocotools.mask as mask_util
from pathlib import Path
import zipfile
from arg_parse import get_args_parser
from tqdm import tqdm
import numpy as np
from detectron2.utils.visualizer import Visualizer
from detectron2.utils.colormap import random_color
from util.server_process import upload_file
CLASSES = ['person', 'giant_panda', 'lizard', 'parrot', 'skateboard', 'sedan', 'ape',
'dog', 'snake', 'monkey', 'hand', 'rabbit', 'duck', 'cat', 'cow', 'fish',
'train', 'horse', 'turtle', 'bear', 'motorbike', 'giraffe', 'leopard',
'fox', 'deer', 'owl', 'surfboard', 'airplane', 'truck', 'zebra', 'tiger',
'elephant', 'snowboard', 'boat', 'shark', 'mouse', 'frog', 'eagle', 'earless_seal',
'tennis_racket']
CLASSES_YVIS_22 = ['airplane', 'bear', 'bird', 'boat', 'car', 'cat', 'cow', 'deer', 'dog', 'duck', 'earless_seal',
'elephant', 'fish', 'flying_disc', 'fox', 'frog', 'giant_panda', 'giraffe', 'horse', 'leopard',
'lizard', 'monkey', 'motorbike', 'mouse', 'parrot', 'person', 'rabbit', 'shark', 'skateboard',
'snake', 'snowboard', 'squirrel', 'surfboard', 'tennis_racket', 'tiger', 'train', 'truck',
'turtle', 'whale', 'zebra']
CLASSES_OVIS = ['Person', 'Bird', 'Cat', 'Dog', 'Horse', 'Sheep', 'Cow', 'Elephant', 'Bear', 'Zebra', 'Giraffe',
'Poultry', 'Giant_panda', 'Lizard', 'Parrot', 'Monkey', 'Rabbit', 'Tiger', 'Fish', 'Turtle', 'Bicycle',
'Motorcycle', 'Airplane', 'Boat', 'Vehical']
transform = T.Compose([
T.Resize(360),
T.ToTensor(),
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
def main(args):
save_vis = False
fast_inference = True
device = torch.device(args.device)
args.train = False
print(device)
# fix the seed for reproducibility
seed = args.seed + utils.get_rank()
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
save_dir = args.initial_save_dir if args.initial_save_dir else args.initial_output_dir
visualization_path = os.path.join(save_dir, "visualization", args.exp_name)
Path(visualization_path).mkdir(parents=True, exist_ok=True)
with torch.no_grad():
model, criterion, postprocessors = build_model(args)
model.to(device)
if args.ckpt_name:
ckpt_list = [os.path.join(args.initial_output_dir, "experiments", args.exp_name + '/' + args.ckpt_name)]
else:
ckpt_list = glob.glob(os.path.join(args.initial_output_dir, "experiments", args.exp_name + '/*.pth'))
ckpt_list.sort(reverse=True)
ckpt_list = ckpt_list[:-1]
videos = json.load(open(args.ann_path, 'rb'))['videos']
vis_num = len(videos)
# now run the model for all the checkpoint
for ckp_id, ckpt in enumerate(ckpt_list):
ckpt_name = os.path.basename(ckpt).split('.')[-2]
print("Processing checkpoint : ", ckpt_name)
# check result dir if already processed
if args.inference_comment:
results_path = os.path.join(save_dir, "results", args.exp_name, args.inference_comment)
analysis_path = os.path.join(args.initial_output_dir, "analysis", args.exp_name, args.inference_comment)
else:
results_path = os.path.join(save_dir, "results", args.exp_name)
analysis_path = os.path.join(args.initial_output_dir, "analysis", args.exp_name)
if os.path.exists(os.path.join(results_path, args.exp_name + "_" + ckpt_name + ".zip")):
print("Results are already processed for this checkpoint")
continue
Path(analysis_path).mkdir(parents=True, exist_ok=True)
state_dict = torch.load(ckpt)['model']
model.load_state_dict(state_dict, strict=False)
model.eval()
folder = args.ytvis_path
result = []
for i in tqdm(range(vis_num)):
id_ = videos[i]['id']
vid_len = videos[i]['length']
file_names = videos[i]['file_names']
video_name_len = 8 if 'ovis' in args.dataset_file else 10
if save_vis: #save vis only for last checkpoint
video_path = f"{visualization_path}/{file_names[0][:video_name_len]}"
Path(video_path).mkdir(parents=True, exist_ok=True)
img_set = []
visualize_img_set = []
vid_range = min(25, vid_len) if (args.debug and 'ovis' in args.dataset_file) else vid_len
for k in range(vid_range):
im = Image.open(os.path.join(folder, file_names[k]))
if save_vis:
visualize_img_set.append(np.array(im))
img_set.append(transform(im).unsqueeze(0).cuda())
img = torch.cat(img_set, 0)
memory_pos = None
model.num_frames = vid_range
outputs = model.inference(img, img.shape[-1], img.shape[-2], memory_pos, fast_inference)
logits = outputs['pred_logits'].squeeze(1)
output_mask = outputs['pred_masks'].squeeze(2).permute(1, 0, 2, 3)
scores = logits.sigmoid().cpu().detach().numpy()
hit_dict = {}
score_dict = {}
num_proposals = 25 * args.increase_proposals if 'ovis' in args.dataset_file else 10 * args.increase_proposals
topkv, indices10 = torch.topk(logits[:3, ...].sigmoid().cpu().detach().flatten(0), k=num_proposals)
indices10 = indices10.tolist()
num_classes = 27 if 'ovis' in args.dataset_file else 42
for idx in indices10:
queryid = (idx % (logits.shape[1] * num_classes)) // num_classes
if args.weighted_category:
topk = torch.topk(logits[:, queryid].sigmoid().cpu().detach().flatten(0), k=(vid_range // 4))
topk_classes = topk.indices % num_classes
hit_dict[queryid] = [torch.mode(topk_classes).values]
score_dict[queryid] = np.mean(
[topk.values[i].item() for i, tc in enumerate(topk_classes) if tc == hit_dict[queryid][0]])
else:
hit_dict[queryid] = [idx % num_classes]
if save_vis:
visualization_masks = [[] for _ in range(vid_range)]
labels = [[] for _ in range(vid_range)]
conf_scores = [[] for _ in range(vid_range)]
for inst_id in hit_dict.keys():
masks = output_mask[inst_id]
pred_masks = F.interpolate(masks[:, None, :, :], (im.size[1], im.size[0]), mode="bilinear")
pred_masks = pred_masks.sigmoid().cpu().detach().numpy() > 0.5
if pred_masks.max() == 0:
print('skip')
continue
for class_id in hit_dict[inst_id]:
category_id = class_id
if args.weighted_category:
score = score_dict[inst_id]
else:
score = scores[:, inst_id, class_id][0]
instance = {'video_id': id_, 'video_name': file_names[0][:video_name_len],
'score': float(score), 'category_id': int(category_id)}
segmentation = []
for n in range(vid_range):
if not args.weighted_category:
score = score[n]
if score < 0.001:
segmentation.append(None)
else:
mask = (pred_masks[n, 0]).astype(np.uint8)
if save_vis:
visualization_masks[n].append(mask)
labels[n].append(int(category_id))
conf_scores[n].append(score)
rle = mask_util.encode(np.array(mask[:, :, np.newaxis], order='F'))[0]
rle["counts"] = rle["counts"].decode("utf-8")
segmentation.append(rle)
instance['segmentations'] = segmentation
result.append(instance)
if save_vis:
num_instances = len(visualization_masks[0])
assigned_colors = [random_color(rgb=True, maximum=1) for _ in range(num_instances)]
for k in range(vid_range):
visualizer = Visualizer(visualize_img_set[k], metadata=None, scale=1.0)
non_zero_mask = [j for j, vm in enumerate(visualization_masks[k]) if vm.sum() > 2]
vis_mask = [visualization_masks[k][i] for i in non_zero_mask]
vis_label = [
CLASSES_OVIS[labels[k][i] - 1] if 'ovis' in args.dataset_file else CLASSES[labels[k][i] - 1] + (
f':{conf_scores[k][i] * 100:.1f}%') for i in non_zero_mask]
vis_color = [assigned_colors[i] for i in non_zero_mask]
vis = visualizer.overlay_instances(labels=vis_label, masks=vis_mask, assigned_colors=vis_color)
vis.save(f"{visualization_path}/{file_names[k]}")
if args.analysis:
analysis_indices = torch.tensor(list(hit_dict.keys()))
hs_analysis = outputs['hs_analysis'][:, analysis_indices, ...]
ref_analysis = outputs['ref_analysis'][:, :,analysis_indices, ...]
init_ref_analysis = outputs['init_ref_analysis'][:, analysis_indices, ...]
np.save(f"{analysis_path}/{videos[i]['file_names'][0].split('/')[0]}_hs_analysis.npy", hs_analysis.cpu().detach().numpy())
np.save(f"{analysis_path}/{videos[i]['file_names'][0].split('/')[0]}_ref_analysis.npy", ref_analysis.cpu().detach().numpy())
np.save(f"{analysis_path}/{videos[i]['file_names'][0].split('/')[0]}_init_ref_analysis.npy", init_ref_analysis.cpu().detach().numpy())
torch.cuda.empty_cache()
Path(results_path).mkdir(parents=True, exist_ok=True)
json_path = f"{results_path}/results.json"
if args.inference_comment:
zip_path = f'{results_path}/{args.exp_name + "_" + args.inference_comment + "_" + ckpt_name}.zip'
else:
zip_path = f'{results_path}/{args.exp_name + "_" + ckpt_name}.zip'
with open(json_path, 'w', encoding='utf-8') as f:
json.dump(result, f)
with zipfile.ZipFile(zip_path, 'w',
compression=zipfile.ZIP_DEFLATED,
compresslevel=9) as zf:
zf.write(json_path, arcname='results.json')
os.remove(json_path)
if args.upload_file:
upload_file(zip_path, args.competition)
if __name__ == '__main__':
print("Initi main")
parser = argparse.ArgumentParser(' inference script', parents=[get_args_parser()])
print("Initi parse")
args = parser.parse_args()
if args.cuda_visible_devices is not None:
os.environ["CUDA_VISIBLE_DEVICES"] = ','.join(map(str, args.cuda_visible_devices))
main(args)