-
Notifications
You must be signed in to change notification settings - Fork 19
/
Copy pathtrain_ninasr.sh
executable file
·77 lines (69 loc) · 3.42 KB
/
train_ninasr.sh
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
#!/bin/bash
train_network () {
arch=$1
scale=$2
echo Training network "${arch}" "x${scale}"
echo python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--log-dir "logs_train/${arch}_x${scale}" --save-checkpoint "${arch}_x${scale}.pt" \
--lr "${learning_rate}" --epochs "${epochs}" --lr-decay-steps $((epochs*2/3)) $((epochs*5/6)) --lr-decay-rate 3 \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4
python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--log-dir "logs_train/${arch}_x${scale}" --save-checkpoint "${arch}_x${scale}.pt" \
--lr "${learning_rate}" --epochs "${epochs}" --lr-decay-steps $((epochs*2/3)) $((epochs*5/6)) --lr-decay-rate 3 \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4 \
--weight-norm
}
train_network_with_pretrained () {
arch=$1
scale=$2
pretrained_scale=$3
echo Pretraining network "${arch}" "x${scale}" from "x${pretrained_scale}"
echo python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--save-checkpoint "${arch}_x${scale}_pre.pt" \
--load-pretrained "${arch}_x${pretrained_scale}_model.pt" \
--lr "${learning_rate}" --epochs 1 --freeze-backbone \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4
python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--save-checkpoint "${arch}_x${scale}_pre.pt" \
--load-pretrained "${arch}_x${pretrained_scale}_model.pt" \
--lr "${learning_rate}" --epochs 1 \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4
echo Training network "${arch}" "x${scale}"
echo python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--log-dir "logs_train/${arch}_x${scale}" --save-checkpoint "${arch}_x${scale}.pt" \
--load-pretrained "${arch}_x${scale}_pre_model.pt" \
--lr "${learning_rate}" --epochs "${epochs}" --lr-decay-steps $((epochs*2/3)) $((epochs*5/6)) --lr-decay-rate 3 \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4
python -m torchsr.train --arch "${arch}" --scale "${scale}" --tune-backend \
--log-dir "logs_train/${arch}_x${scale}" --save-checkpoint "${arch}_x${scale}.pt" \
--load-pretrained "${arch}_x${scale}_pre_model.pt" \
--lr "${learning_rate}" --epochs "${epochs}" --lr-decay-steps $((epochs*2/3)) $((epochs*5/6)) --lr-decay-rate 3 \
--patch-size-train $(( (patch_size+1) * scale)) --shave-border "${scale}" --replication-pad 4 \
--weight-norm
}
epochs=300
patch_size=48
learning_rate=0.001
# NinaSR-B0
train_network ninasr_b0 2
train_network ninasr_b0 3
train_network ninasr_b0 4
train_network ninasr_b0 8
epochs=500
patch_size=48
learning_rate=0.0003
# NinaSR-B1
train_network ninasr_b1 2
train_network_with_pretrained ninasr_b1 3 2
train_network_with_pretrained ninasr_b1 4 3
train_network_with_pretrained ninasr_b1 8 4
train_network_with_pretrained ninasr_b1 2 4
epochs=1000
patch_size=48
learning_rate=0.0001
# NinaSR-B2
train_network ninasr_b2 2
train_network_with_pretrained ninasr_b2 3 2
train_network_with_pretrained ninasr_b2 4 3
train_network_with_pretrained ninasr_b2 8 4
train_network_with_pretrained ninasr_b2 2 4