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!
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!