Not working on pictures outside dataset
I have tried to use it on sevaral images. It only makes the shadow grey.
How to make it more reliable on other images?
Hi @yongsuH , our method is very robustness to real shadow images as you can see in our video demo https://www.youtube.com/watch?v=WvWBoOZR208. Your results could be lead by incorrect configuration.
You can try our live demo at huggingface space: https://huggingface.co/spaces/MKFMIKU/Instance-Shadow-Removal
Hi @yongsuH , our method is very robustness to real shadow images as you can see in our video demo https://www.youtube.com/watch?v=WvWBoOZR208. Your results could be lead by incorrect configuration.
You can try our live demo at huggingface space: https://huggingface.co/spaces/MKFMIKU/Instance-Shadow-Removal
I'm using default settings from shadow_aistd_test.py and it gives good results on AISTD images.
Can you try the code and ckpt here: https://huggingface.co/spaces/MKFMIKU/Instance-Shadow-Removal/tree/main ?
Can you try the code and ckpt here: https://huggingface.co/spaces/MKFMIKU/Instance-Shadow-Removal/tree/main ?
I edited the app.py to run locally, here's the result with dkernel = 11, diffusion_steps = 50
It performs pretty much the same as the code from this git repo.
@yongsuH That's interesting. Can you share the original input so I can take a try?
Sure.