-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_detr_ds.py
192 lines (169 loc) · 7.18 KB
/
train_detr_ds.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
import os, torch, deepspeed, random, argparse
import numpy as np
import dataset.transforms as T
from model.backbone import ResNetBackbone
from model.transformer import TransformerBitLinear
from model.detr import DETR, SetCriterion
from model.matcher import HungarianMatcher
from dataset.coco import CocoDetection, collate_fn
def get_args_parser():
parser = argparse.ArgumentParser('Set transformer detector', add_help=False)
parser.add_argument('--lr', default=1e-4, type=float)
parser.add_argument('--lr_backbone', default=1e-5, type=float)
parser.add_argument('--batch_size', default=16, type=int)
parser.add_argument('--weight_decay', default=1e-4, type=float)
parser.add_argument('--epochs', default=300, type=int)
parser.add_argument('--lr_drop', default=200, type=int)
parser.add_argument('--clip_max_norm', default=0.1, type=float,
help='gradient clipping max norm')
# Backbone
parser.add_argument('--backbone', default='resnet50', type=str,
help="Name of the convolutional backbone to use")
# Transformer
parser.add_argument('--enc_layers', default=6, type=int,
help="Number of encoding layers in the transformer")
parser.add_argument('--dec_layers', default=6, type=int,
help="Number of decoding layers in the transformer")
parser.add_argument('--dim_feedforward', default=2048, type=int,
help="Intermediate size of the feedforward layers in the transformer blocks")
parser.add_argument('--hidden_dim', default=256, type=int,
help="Size of the embeddings (dimension of the transformer)")
parser.add_argument('--dropout', default=0.1, type=float,
help="Dropout applied in the transformer")
parser.add_argument('--nheads', default=8, type=int,
help="Number of attention heads inside the transformer's attentions")
parser.add_argument('--num_queries', default=100, type=int,
help="Number of query slots")
# Matcher
parser.add_argument('--cost_class', default=1, type=float,
help="Class coefficient in the matching cost")
parser.add_argument('--cost_bbox', default=5, type=float,
help="L1 box coefficient in the matching cost")
parser.add_argument('--cost_giou', default=2, type=float,
help="giou box coefficient in the matching cost")
# Loss coefficients
parser.add_argument('--dice_loss_coef', default=1, type=float)
parser.add_argument('--bbox_loss_coef', default=5, type=float)
parser.add_argument('--giou_loss_coef', default=2, type=float)
parser.add_argument('--eos_coef', default=0.1, type=float,
help="Relative classification weight of the no-object class")
# Dataset parameters
parser.add_argument('--dataset_file', default='coco')
parser.add_argument('--coco_path', type=str)
parser.add_argument('--num_classes', default=91, type=int)
# Others
parser.add_argument('--output_dir', default='',
help='path where to save, empty for no saving')
parser.add_argument('--device', default='cuda',
help='device to use for training / testing')
parser.add_argument('--seed', default=42, type=int)
parser.add_argument('--resume', default='', help='resume from checkpoint')
parser.add_argument('--start_epoch', default=0, type=int, metavar='N',
help='start epoch')
parser.add_argument('--eval', action='store_true')
parser.add_argument('--num_workers', default=2, type=int)
# Distributed training parameters
parser.add_argument('--world_size', default=1, type=int,
help='number of distributed processes')
parser.add_argument('--local_rank', default=-1, type=int,
help='parameters used by deepspeed')
# Segmentation
parser.add_argument('--masks', action='store_true',
help="Train segmentation head if the flag is provided")
return parser
def main(args):
deepspeed.init_distributed()
ds_config = {
"train_batch_size": args.batch_size,
"optimizer": {
"type": "Adam",
"params": {
"lr": 1e-5
}
},
"fp16": {
"enabled": True,
"auto_cast": True,
"loss_scale": 0,
"initial_scale_power": 11,
},
"zero_optimization": {
"stage": 1,
"reduce_bucket_size": 5e8,
},
"tensorboard": {
"enabled": True,
"output_path": "run/",
"job_name": "train_detr_bitlinear_pred",
},
"comms_logger": {
"enabled": True,
"verbose": False,
"prof_all": True,
"debug": False
}
}
rank = int(os.getenv("LOCAL_RANK", "0"))
seed = args.seed + rank
torch.manual_seed(seed)
np.random.seed(seed)
random.seed(seed)
# TODO: replace with Albumentation
scales = [480, 512, 544, 576, 608, 640, 672, 704, 736, 768, 800]
transform_train = T.Compose([
# augumentation
T.RandomHorizontalFlip(),
T.RandomSelect(
T.RandomResize(scales, max_size=1333),
T.Compose([
T.RandomResize([400, 500, 600]),
T.RandomSizeCrop(384, 600),
T.RandomResize(scales, max_size=1333),
])
),
# normalize
T.Compose([
T.ToTensor(),
T.Normalize([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
])
])
# avoid downloading the same file at the same time during model initialization
if rank != 0: torch.distributed.barrier()
dataset_train = CocoDetection(args.coco_path + "/train2017", args.coco_path + "/annotations/instances_train2017.json", transform_train)
# TODO: use a ViT based backbone
backbone = ResNetBackbone()
transformer = TransformerBitLinear(args.hidden_dim, args.nheads, args.enc_layers, args.dec_layers, args.dim_feedforward, args.dropout)
model = DETR(backbone=backbone, transformer=transformer, num_classes=args.num_classes, num_queries=args.num_queries)
matcher = HungarianMatcher(args.cost_class, args.cost_bbox, args.cost_giou)
# TODO: let criterion return all 3 losses and apply the weight at the end before backprop
criterion = SetCriterion(args.num_classes, matcher, args.eos_coef, weight=(args.dice_loss_coef, args.bbox_loss_coef, args.giou_loss_coef))
if rank == 0: torch.distributed.barrier()
# TODO: calculate the number of tenary parameters for transformers
n_parameters = sum(p.numel() for p in model.parameters() if p.requires_grad)
print('number of params:', n_parameters)
# deepseed initialization
model_engine, optimizer, data_loader_train, _ = deepspeed.initialize(model=model, model_parameters=model.parameters(),
training_data=dataset_train, collate_fn=collate_fn,
config=ds_config)
device = torch.device("cuda", rank)
model.to(device)
criterion.to(device)
for epoch in range(args.start_epoch, args.epochs):
model.train()
criterion.train()
for samples, masks, targets in data_loader_train:
optimizer.zero_grad()
samples = samples.to(device)
masks = masks.to(device)
# TODO: separate them into targets_class and target_bboxes
targets = [{k: v.to(device) for k, v in t.items()} for t in targets]
with torch.autocast(device_type="cuda", dtype=torch.float16):
outputs_logits, outputs_boxes = model_engine(samples, masks)
loss = criterion(outputs_logits, outputs_boxes, targets)
model_engine.backward(loss)
model_engine.step()
# TODO: add eval on rank=0
if __name__ == '__main__':
parser = argparse.ArgumentParser('DETR training and evaluation script', parents=[get_args_parser()])
args = parser.parse_args()
main(args)