Hugging Face's logo

Yuanzhi
/
OFTSR

Model card Files Community
OFTSR
File size: 4,988 Bytes 
5a72721
 
 
 
90a9b04
5a72721
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37e7189
5a72721
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37e7189
5a72721
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37e7189
5a72721
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
37e7189
5a72721
 
 
 
 
 
 
 
 
 
 
 
 
 
90a9b04
5a72721
 
 
 
 
 
 
 
 
 
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
 
---
license: apache-2.0
---

## Checkpoints for [OFTSR](https://github.com/yuanzhi-zhu/OFTSR)

```bash
cd OFTSR
huggingface-cli download Yuanzhi/OFTSR --local-dir ckpts
```
put data at ./val_data

########################## FFHQ Generation ##########################
```bash
python sample_fm.py --opt configs/ir_fm_ffhq.yml \
        --overrides \
        sample.pre_train_model=ckpts/guided_unet-sigma1.0-no_cond-bs128-loss_l2-lr0.0001-FFHQ-checkpoint_26.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        ir.sigma_pertubation=1.0 \
        fm_model.use_cond=false
```

########################## FFHQ Noiseless Restoration ##########################

#--------------------- multi-step ---------------------
```bash
python sample_fm.py --opt configs/ir_fm_ffhq.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.1-bs32-loss_l1-lr0.0001-FFHQ-checkpoint_10.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.1 \
        fm_model.use_cond=true
```

#--------------------- one-step ---------------------
```bash
python sample_fm.py --opt configs/dis_fm_ffhq.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.1-bs32-loss_l1-lr1e-05-distil-v_boot-solver_rk2_0.5-dt0.05-w_distil_1.0-w_bound_0.1-w_align_0.01-FFHQ_DIS-checkpoint_2.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.1 \
        fm_model.use_cond=true
        sample.one_step_t=0.99
```

########################## FFHQ Noisy Restoration ##########################

#--------------------- multi-step ---------------------
```bash
python sample_fm.py --opt configs/ir_fm_ffhq.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_avp-sf4-sigmay_0.05-guided_unet-sigma0.5-bs128-loss_l1-lr0.0001-FFHQ-checkpoint_23.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0.05 \
        ir.sigma_pertubation=0.5 \
        fm_model.use_cond=true
```

#--------------------- one-step ---------------------
```bash
python sample_fm.py --opt configs/dis_fm_ffhq.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_avp-sf4-sigmay_0.05-guided_unet-sigma0.5-bs32-loss_l1-lr2e-05-distil-v_boot-solver_rk2_0.5-dt0.05-w_distil_1.0-w_bound_0.1-w_align_0.01-FFHQ_DIS-checkpoint_24.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0.05 \
        ir.sigma_pertubation=0.5 \
        fm_model.use_cond=true
        sample.one_step_t=0.99
```

########################## DIV2K Noiseless Restoration ##########################

#--------------------- multi-step ---------------------
```bash
python sample_fm.py --opt configs/ir_fm_DIV2K.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.2-bs128-loss_l1-lr1e-05-DIV2K-checkpoint_4.pth \
        dataset.val_path='./val_data/DIV2K_valid_HR' \
        sample.num_sample=100 \
        sample.psnr_batch_size=1 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.2 \
        fm_model.use_cond=true
```

#--------------------- one-step ---------------------
```bash
python sample_fm.py --opt configs/dis_fm_DIV2K.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.2-bs32-loss_l1-lr1e-05-distil-v_boot-solver_rk2_0.5-dt0.05-w_distil_1.0-w_bound_0.1-w_align_0.01-DIV2K_DIS-checkpoint_5.pth \
        dataset.val_path='./val_data/ffhq_val_100' \
        sample.num_sample=100 \
        sample.psnr_batch_size=1 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.2 \
        fm_model.use_cond=true
        sample.one_step_t=0.99
```



########################## ImageNet Noiseless Restoration ##########################

#--------------------- multi-step ---------------------
```bash
python sample_fm.py --opt configs/ir_fm_imagenet.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.2-bs32-loss_l1-lr1e-04-ImageNet-checkpoint_10.pth \
        dataset.val_path='./val_data/imagenet_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.2 \
        fm_model.use_cond=true
```

#--------------------- one-step ---------------------
```bash
python sample_fm.py --opt configs/dis_fm_imagenet.yml \
        --overrides \
        sample.pre_train_model=ckpts/sr_bicubic-sf4-guided_unet-sigma0.2-bs8-loss_l1-lr1e-04-distil-v_boot-solver_rk2_0.5-dt0.05-w_distil_1.0-w_bound_0.1-w_align_0.01-ImageNet_DIS-checkpoint_10.pth \
        dataset.val_path='./val_data/imagenet_val_100' \
        sample.num_sample=100 \
        ir.sigma_y=0. \
        ir.sigma_pertubation=0.2 \
        fm_model.use_cond=true
        sample.one_step_t=0.99
```