Questions regarding LPN training and composition loss #7
Description
Thank you for this great job! I have a few questions:
-
Is it part of your plan to release the code for training LPN? I couldn't find it in your current repository.
-
When training SHM, are the weights of LPN not frozen? Does this mean that during SHM training, the weights of LPN will also be updated?
-
How would training SHM without a pretrained LPN affect the performance?
-
Could you please provide more details on how you generated the fgr and bgr images for the HHM50K dataset? Were they created using a project like [https://github.com/MarcoForte/closed-form-matting]
-
I noticed that you used the composition loss in your code and set the composition_loss_weight to 0.5. However, I couldn't find any information about the composition loss in your paper. Have you conducted any ablation studies on the composition loss?
I am looking forward to your response. Thank you very much!