Pretrained models can be downloaded from Google Drive. You can put them in experiments/pretrained_models folder.
- SRResNet_bicx4_in3nf64nb16.pth: SRResNet x4 model, trained on DIV2K, w/o BN, bicubic downsampling.
- SRGAN_bicx4_303_505.pth: SRGAN x4 model, trained on DIV2K, w/o BN, bicubic downsampling.
- segmentation_OST_bic.pth: segmentation model for bicubiced images, outdoor scenes.
- sft_net_torch.pth: torch version of SFTGAN model.
- sft_net_ini.pth: initialized SFTGAN model, initializing the sr generator with SRGAN_bicx4_303_505 parameters.
- SFTGAN_bicx4_noBN_OST_bg.pth: SFTGAN model, trained on OST dataset and use DIV2K as background images, w/o BN, bicubic downsampling.
Through experiments, we found that
- no batch normalization
- residual block style: Conv-ReLU-Conv
are the best network settings.
SRResNet_bicx4_in3nf64nb16.pth is provided here and other pretrained models can be downloaded from Google Drive.
| Model | Scale | Channel | DIV2K2 | Set5 | Set14 | BSD100 | Urban100 |
|---|---|---|---|---|---|---|---|
| SRResNet_bicx2_in3nf64nb161 | 2 | RGB | 34.7203 | 35.835 | 31.643 | ||
| 36.1433 | 37.947 | 33.682 | |||||
| SRResNet_bicx3_in3nf64nb16 | 3 | RGB | 31.019 | 32.442 | 28.499 | ||
| 32.449 | 34.428 | 30.371 | |||||
| SRResNet_bicx4_in3nf64nb16 | 4 | RGB | 29.051 | 30.278 | 26.853 | ||
| 30.486 | 32.180 | 28.645 | |||||
| SRResNet_bicx8_in3nf64nb16 | 8 | RGB | 25.429 | 25.357 | 23.348 | ||
| 26.885 | 27.070 | 24.996 | |||||
| SRResNet_bicx2_in1nf64nb16 | 2 | Y | 35.870 | 37.864 | 33.581 | ||
| SRResNet_bicx3_in1nf64nb16 | 3 | Y | 32.182 | 34.263 | 30.186 | ||
| SRResNet_bicx4_in1nf64nb16 | 4 | Y | 30.224 | 32.0384 | 28.494 | ||
| SRResNet_bicx8_in1nf64nb16 | 8 | Y | 26.660 | 26.621 | 24.804 |
1 bic: MATLAB bicubic downsampling; in3: input has 3 channels; nf64: 64 feature maps; nb16: 16 residual blocks.
2 DIV2K 0801 ~ 0900 validation images.
3 The first row is evaluated on RGB channels, while the secone row is evaluated on Y channel (of YCbCr).
4 (31.901, 29.711)