On GeForce 30 series it's not work

[Mihailllll]

On GeForce 30 series I get error: nvcc fatal : Unsupported gpu architecture 'compute_86'

[dueToLife]

Hi，I successfully run this repo on GeForce 3090. I have met similar problem of unsupported gpu. I think env in requirement.txt may be out of date. SO I change env into pytorch 1.8+cu11.1 , and then demos run ok.

[Mihailllll]

I rebuild cuda and get new error:

  File "/PIXIE/pixielib/utils/tensor_cropper.py", line 78, in crop_tensor
    cropped_image = warp_affine(
TypeError: warp_affine() got an unexpected keyword argument 'flags'

Code in project: # warp images cropped_image = warp_affine( image, dst_trans_src[:, :2, :], (crop_size, crop_size), flags=interpolation, align_corners=align_corners)

How fix this error? I have torch 1.11 and cuda 11.6 torch.cuda.is_available() return true

[dueToLife]

That‘s because flags is deprecated in new version kornia. Just modify /PIXIE/pixielib/utils/tensor_cropper.py, line 78, flags=... into mode=... like: cropped_image = warp_affine( image, dst_trans_src[:, :2, :], (crop_size, crop_size), mode=interpolation, align_corners=align_corners)

[Mihailllll]

Thanks. But now i get error:

File "/home/owner/PythonProjects/3dBodyFromImage/4/PIXIE/pixielib/utils/renderer.py", line 84, in forward pixel_vals = (bary_coords[..., None] * pixel_face_vals).sum(dim=-2) RuntimeError: CUDA out of memory. Tried to allocate 144.00 MiB (GPU 0; 3.82 GiB total capacity; 1.63 GiB already allocated; 139.00 MiB free; 1.87 GiB reserved in total by PyTorch) If reserved memory is >> al

How fix this error?

[dueToLife]

i didn't meet this. Error information means your GPU run out of memory. Make sure you have enough GPU memory. Good Luck!

[Mihailllll]

I use one image and it is work, thanks!