hjsim / koalanet Goto Github PK

Thank you so much for this amazing work. And congratulations on the research paper accepted at CVPR.

I want to evaluate your models on the Set 5 dataset. However, when I ran the model (4X) with the default test_patch argument [1,1], the results were images with total noise. I don't know for sure what's the issue, but I tried changing the default test_patch argument to [2,2] and got the following results. It is a little better than totally noisy images. Changing test_patch size is affecting the resulting super-resolution image, so what should be the ideal value for the test_patch parameter?

As you can see the results do not look good. What am I missing here?

It will mean a lot if I can get any leads about what can be going wrong here.

Thank you in advance!

hi thanks for your great work...can you please let me know what is the main difference between kernel and filter.

Hi,

Thanks for the code.

I tried to test the pre-trained models on the Set5 dataset you provided running:
python main.py --phase 'test' --test_data_path 'Set5/LR/X4/imgs' --test_label_path 'Set5/HR' --test_ckpt_pat ./upsampling_network_x4 --factor 4.

However, the results have grid artefacts like this one:

I am using TensorFlow 1.13 like mentionned in the requirements.

Is there something I missed in the params ?

Thanks,
Charles

Could you help me?