Fork of Lornatang/ESPCN-PyTorch

Differences between original repository and fork:

• Compatibility with PyTorch >=2.4. (🔥)
• Original pretrained models and converted ONNX models from GitHub releases page. (🔥)
• Model conversion to ONNX format using the export.py file. (🔥)
• Installation with updated requirements.txt file.
• The following deprecations has been fixed:
  • FutureWarning: You are using 'torch.load' with 'weights_only=False'.

Installation

pip install -r requirements.txt

Pretrained models

• Download links:

Name	Model Size (MB)	Link	SHA-256
ESPCN_x2	0.1 0.1	PyTorch, ONNX	67321a870da341c15a92f5dcea31bcb21b7fa30165b0ac445662d4364048595b 6c806349be7f963f3d0895050a771a5de15e9950361e2a2f92624e4b1f675044
ESPCN_x3	0.1 0.1	PyTorch, ONNX	242ab640f79fa1e005f5d495debf6c1e385209c8b81f14f5692ba7ed51ec2f5d 77aea5de0ba9628566ef9519de06a18ff5c25b32b0743d3e49a808c539c5445f
ESPCN_x4	0.1 0.1	PyTorch, ONNX	756564c1b4103cad1170479bbce665b2b58163444d5170dcc1db37ba143694f8 2b421d032afd5737c8cc0f1145214d78c35dbc46e3c7cb45913da80304b8aa8e

• Evaluation results:

Name	Scale	Set5 (PSNR)	Set14 (PSNR)
ESPCN_x2	2	36.64	32.35
ESPCN_x3	3	32.55	29.20
ESPCN_x4	4	30.26	27.41

Inference

python inference.py --model_arch_name espcn_x2 --upscale_factor 2 --model_weights_path espcn-x2.pth.tar --inputs_path figure/comic.png --output_path figure/sr_comic_x2.png
python inference.py --model_arch_name espcn_x3 --upscale_factor 3 --model_weights_path espcn-x3.pth.tar --inputs_path figure/comic.png --output_path figure/sr_comic_x3.png
python inference.py --model_arch_name espcn_x4 --upscale_factor 4 --model_weights_path espcn-x4.pth.tar --inputs_path figure/comic.png --output_path figure/sr_comic_x4.png

Export to ONNX format

pip install onnx

python export.py --model_arch_name espcn_x2 --model_weights_path espcn-x2.pth.tar
python export.py --model_arch_name espcn_x3 --model_weights_path espcn-x3.pth.tar
python export.py --model_arch_name espcn_x4 --model_weights_path espcn-x4.pth.tar