forked from southnx/ACoLP
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
183 lines (160 loc) · 7.37 KB
/
main.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
import torch
import torch.nn as nn
import torch.nn.functional as F
import torchvision.ops.roi_align
import torch.optim as optim
from torch.optim.lr_scheduler import StepLR, ExponentialLR
from torchvision import transforms
import torch.distributed as dist
from torch.nn.parallel import DistributedDataParallel as DDP
from torch.utils.data import DataLoader
from torch.utils.data.distributed import DistributedSampler
from timm.models.layers import trunc_normal_
from torchvision.models.detection import (fasterrcnn_resnet50_fpn,
fasterrcnn_resnet50_fpn_v2,
FasterRCNN_ResNet50_FPN_Weights,
FasterRCNN_ResNet50_FPN_V2_Weights)
import numpy as np
import os, tqdm
import gc
import time
import random
import argparse
from datetime import datetime
from models.args import get_args
import utils.factory as utils
from datasets.VidHOI_dataset import VidHOI_keyframe_Dataset
from models.ACoLP import ACoLP
def init_weights(m):
if isinstance(m, nn.Linear):
trunc_normal_(m.weight, std = .02)
if isinstance(m, nn.Linear) and m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.LayerNorm):
nn.init.constant_(m.bias, 0)
nn.init.constant_(m.weight, 1.0)
def main():
args = get_args()
args.world_size = args.gpus * args.nodes
device = torch.device('cuda', args.local_rank)
BATCH_SIZE = args.batch_size
# device = torch.device(args.device)
exp_time = datetime.now().strftime("%Y-%m-%d_%H:%M")
if args.local_rank == 0:
print("#" * 80)
# print("# - Experiment: {}".format(exp_descp))
print("# - Experiment start on: {}".format(exp_time))
print("# - {}".format(args))
print("#" * 80)
seed = args.seed + utils.get_rank()
torch.manual_seed(seed)
torch.cuda.manual_seed(seed)
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
np.random.seed(seed)
torch.cuda.manual_seed_all(seed)
random.seed(seed)
gc.collect() # empty RAM
torch.cuda.empty_cache()
torch.cuda.set_device(args.local_rank)
ROOT_PATH = '/home/data/Dataset/Vid_HOI'
JSON_ANNO_TRAIN = '/home/data/Dataset/Vid_HOI/train_frame_annots.json'
JSON_ANNO_VAL = '/home/data/Dataset/VidHOI_STHOI_paper_code/slowfast/datasets/vidor-github/val_frame_annots.json'
weights = FasterRCNN_ResNet50_FPN_V2_Weights.COCO_V1.DEFAULT
train_dst = VidHOI_keyframe_Dataset(
root_dir = ROOT_PATH,
keyframe_folder = 'small_keyframes_train',
annotation = JSON_ANNO_TRAIN,
frames_per_clip = 8,
transform=transforms.Compose(
[transforms.Resize(args.re_size),
# transforms.RandomHorizontalFlip(p=0.5),
# transforms.RandomCrop((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])]
))
val_dst = VidHOI_keyframe_Dataset(
root_dir = ROOT_PATH,
keyframe_folder = 'small_keyframes_test',
annotation = JSON_ANNO_VAL,
frames_per_clip = 8,
transform=transforms.Compose(
[transforms.Resize(args.re_size),
# transforms.RandomCrop((224, 224)),
transforms.ToTensor(),
transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])]
))
if args.local_rank == 0:
print("Train size:", len(train_dst)) # Train size: 27087
print("Val size:", len(val_dst)) # Val size: 3216
# model = LangSupVidHOI()
model = ACoLP()
if args.distri:
model.apply(init_weights)
model = torch.nn.SyncBatchNorm.convert_sync_batchnorm(model)
model = DDP(model.to(device), device_ids=[args.local_rank], output_device=args.local_rank, find_unused_parameters=True)
train_sampler = DistributedSampler(train_dst, num_replicas=None, rank=None, shuffle=True, seed=42, drop_last=False)
train_batch_sampler = torch.utils.data.BatchSampler(
train_sampler, batch_size=BATCH_SIZE // 6, drop_last=True)
train_dataloader = DataLoader(
train_dst, num_workers=4, batch_sampler=train_batch_sampler)
else:
# model.apply(init_weights).to(device)
model.apply(init_weights).cuda()
train_dataloader = DataLoader(train_dst, batch_size=BATCH_SIZE // 6, num_workers=4,
shuffle=True, drop_last=False, collate_fn=lambda x: x)
print("length of dataloader: ", len(train_dataloader)) # len(train_dst) / batch_size
# for data in tqdm(total_dataloader, position=0, decs = 'load train dataset'):
# pass
# total_dataloader.dataset.set_use_cache(use_cache=True)
criterion = nn.BCEWithLogitsLoss().to(device)
# optimizer = optim.Adam(model.parameters(), lr=args.lr * args.lr_scale)
# exp_lr_scheduler = StepLR(optimizer, step_size=args.step_size, gamma=0.9)
for epoch in range(0, args.epochs):
torch.cuda.empty_cache()
total_step = len(train_dataloader)
if args.local_rank == 0:
print("Epoch:{}".format(epoch))
print("total step: ", total_step)
# train_sampler.set_epoch(epoch)
epoch_since = time.time()
# hs = open(os.path.join(save_path, "output.txt"), "a")
# if args.local_rank == 0:
# hs.write("Mode: {} \t".format(args.mode))
cunt_train = 0
for i, sample_batched in enumerate(train_dataloader):
cunt_train += 1
# print(sample_batched[0]['clip_labels'])
# inputs, labels = torch.tensor(sample_batched[0]['clip_frames']).to(device), torch.tensor(sample_batched[0]['clip_labels']).to(device)
input_list = []
ori_img_list = []
bbox_ratio_list = []
label_list = []
frame_name_list = []
for ii in range(len(sample_batched[0]['clip_frames'])):
# print("sampled batch: ", sample_batched[0]['keyframe_names'])
inputs = sample_batched[0]['clip_frames'][ii].requires_grad_(True).to(device)
input_list.append(inputs)
ori_clip_frames = sample_batched[0]['original_frames'][ii].to(device)
ori_img_list.append(ori_clip_frames)
ori_img_size = sample_batched[0]['original_frame_size'][ii]
bbox_ratio_x = args.re_size[0] / ori_img_size[1]
bbox_ratio_y = args.re_size[1] / ori_img_size[2]
bbox_ratio_list.append((bbox_ratio_x, bbox_ratio_y))
labels = torch.tensor(sample_batched[0]['clip_labels'][ii]).to(device)
label_list.append(labels)
frame_name = sample_batched[0]['keyframe_names'][ii]
frame_name_list.append(frame_name)
# print("inputs shape: ", inputs.shape) # torch.Size([3, 224, 224])
# edge_fea = model(inputs, labels)
print("frame: ", frame_name)
torch.cuda.synchronize()
print("-" * 50)
print("cunt_train: ", cunt_train)
print("bbox ratio: ", bbox_ratio_list)
# print("input_list: ", input_list)
# print("HOI label_list: ", label_list)
# fea = model(input_list)
fea = model(ori_img_list, input_list, frame_name_list, bbox_ratio_list)
if __name__=='__main__':
main()