The super resolution result of the given test image is blur. No errors are reported. I wonder where I may go wrong?

cswry / SeeSR

[CVPR2024] SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution

Apache License 2.0

331 stars 14 forks source link

Closed CXNing closed 2 months ago

CXNing commented 4 months ago

微信截图_20240229115633

cswry commented 4 months ago

Hello, can you provide the resolution of input image?

CXNing commented 4 months ago

128*128, it's the image in preset/datasets/test_datasets

CXNing commented 4 months ago

So what's the possible reason? The wrong checkpoint? But the model is loaded without any errors.

cswry commented 4 months ago

So what's the possible reason? The wrong checkpoint? But the model is loaded without any errors.

It's a strange problem. The demo runs the same case as yours and it work normally.

Have you ever run this demo with test_seesr.py rather than the grdio app?

CXNing commented 4 months ago

Yes, the result is the same. Strange, I re-downloaded the model and the problem still exists.

CXNing commented 4 months ago

The tag is “animal, pink, stone, salamander, None,clean, high-resolution, 8k," right? Does the None tag influences the result?

cswry commented 3 months ago

The tag is “animal, pink, stone, salamander, None,clean, high-resolution, 8k," right? Does the None tag influences the result?

What would happen if you delete 'none'?

Normally, DAPE won't predict 'none', I guess you mistakenly set the text in the user prompt.

noir94 commented 3 months ago

I have the same results.. Have you solved this problem?

CXNing commented 2 months ago

I have the same results.. Have you solved this problem?

CXNing commented 2 months ago

Sorry, there is a problem with downloading my SD2-base model.