why

[ld-xy]

Why is your code on github different from the code on huggingface? ? ? ?

[niki-amini-naieni]

The Hugging Face App is meant for simple single-image inference and does not have adaptive cropping or the SAM test-time normalization applied. On the other hand, the GitHub code has these additions. The reason for this is that the app is a simpler stripped-down version of the code intended for users to try out custom images, etc... On the other hand, the GitHub code is intended to show how to exactly reproduce the results from the paper. The Hugging Face app code is based on the GitHub code. The pretrained model is the same for both codebases though. In short, the Hugging Face code is a simplified version of the GitHub code, but both repositories use the same pretrained CountGD model.