xhlbetter issues

Repositories
Issues
Comments

Results 3 issues of


                                            xhlbetter

svd_cuda: the updating process of SBDSDC did not converge (error: 2)

When I train the model from scratch, sometimes I encounter such an error: ![image](https://github.com/MCG-NJU/SADRNet/assets/105650177/8eb365cb-7262-4cc1-8475-76e4b418bf64) nan in H ![image](https://github.com/MCG-NJU/SADRNet/assets/105650177/3f7b73e0-fc55-4b30-af7b-00a0354d6e3b)

PIL.UnidentifiedImageError

![image](https://github.com/MCG-NJU/SADRNet/assets/105650177/277c7d28-81a3-4f0d-92ee-7b00b275ce56)

training loss

3DMM's shape_para, exp_para, and Tm values are very large, causing a large loss in the early stages of training. How can this be addressed?