Open ZhimingNJ opened 8 months ago
@ZhimingNJ Hi, I think it's because we train the model with a fixed camera distance, where the input image should always have a border to the edge. Recentering is basically to make sure the user input is similar to our training data.
Thank you for your generosity to release your source code, this is a great work!
If skipping recenter() in infer.py, reconstruction result is worse. Why does it matter for reconstruction?
thanks