-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathvggss_sync_contrast.yaml
70 lines (66 loc) · 1.44 KB
/
vggss_sync_contrast.yaml
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
exp:
output_dir: "exps/avsync/vggss_sync_contrast"
log_file: "exps/avsync/vggss_sync_contrast/train.log"
log_with: "wandb"
seed: 123
model:
audio_encoder:
name: AudioConv2DNet
pretrained: True
video_encoder:
name: VideoR2Plus1DNet
pretrained: True
head:
name: FCHead
dim: 512
out_dim: 1
dropout: 0.0
tau: 0.1
train:
batch_size: 4
log_steps: 10
dataset:
data_root: "./datasets/VGGSS/videos"
example_list_path: "./datasets/VGGSS/train.txt"
mode: "train"
image_size: 224
video_fps: 6
video_num_frames: 12
audio_sample_rate: 16000
randflip: True
shift_time: 0.2
num_clips: 21
sampling_type: "random-compact"
test:
batch_size: 8
test_steps: 2500
dataset:
data_root: "./datasets/VGGSS/videos"
example_list_path: "./datasets/VGGSS/test.txt"
mode: "test"
image_size: 224
video_fps: 6
video_num_frames: 12
audio_sample_rate: 16000
randflip: False
shift_time: 0.2
num_clips: 21
sampling_type: "uniform"
# optimizer
optim:
max_train_steps: 350000
learning_rate: 2e-4
scale_lr: False
lr_scheduler: "constant_with_warmup"
lr_warmup_steps: 100
adam_beta1: 0.9
adam_beta2: 0.999
adam_weight_decay: 1e-2
adam_epsilon: 1e-08
max_grad_norm: 1.0
gradient_accumulation_steps: 1
checkpointing_steps: 2500
checkpointing_milestones: 40000
resume_from_checkpoint: "latest"
mixed_precision: "fp16"
use_8bit_adam: False