sunniesuhyoung / dst Goto Github PK

colab fetches a folder in your google drive instead of cloning the repo which is a bit different than this repo in getting the nbb points

Thank you for sharing your amazing work!!

I'm trying to run the code using my own data, but singularity error happens every time.

I've checked that zwfcrazy raised the same issue. In response to this, RazvanRotaru said deleting the NBB and re-downloading it would help.
However, deleting NBB doesn't really help.

Could you help me to solve this problem?

Traceback (most recent call last):
File "main.py", line 115, in
device=device)
File "/root/DST/styletransfer.py", line 143, in DST
new_im, content_im_warp, warp_field = apply_warp(new_im, [src_Kpts_aug], [dst_Kpts_aug], device, sharp=sharp_warp, im2=content_im_warp)
File "/root/DST/warp.py", line 432, in apply_warp
new_im, new_im2, warp_field = image_warp(new_im, dst, dst-src, device, sharp=sharp, img2=new_im2)
File "/root/DST/warp.py", line 86, in image_warp
w, v = solve_interpolation(dst_pts, flow_pts, device)
File "/root/DST/warp.py", line 226, in solve_interpolation
w_v, _ = torch.solve(rhs, lhs)
RuntimeError: solve_cuda: U(459,459) is zero, singular U.

Hello, does this model works only for single image translation or it is also possible to train for multiple images? I mean I tried to reproduce your work with my own datasets but I couldn't train for a large number of images it requires only one image at a time.

Hello, Amazing work!
Can you share some trained weights for evaluation and testing purpose?

Hi,
Thank you for open sourcing such a great work.

Can you tell me the inference time of the model(end to end , including finding NBB and everything) ?
I would have checked it but seems like not all files have been added yet.

Thanks.

Sometimes singularity happens when calling solve_interpolation.

Traceback (most recent call last):
File "main.py", line 115, in
device=device)
File "/root/DST/styletransfer.py", line 144, in DST
new_im, content_im_warp, warp_field = apply_warp(new_im, [src_Kpts_aug], [dst_Kpts_aug], device, sharp=sharp_warp, im2=content_im_warp)
File "/root/DST/warp.py", line 432, in apply_warp
new_im, new_im2, warp_field = image_warp(new_im, dst, dst-src, device, sharp=sharp, img2=new_im2)
File "/root/DST/warp.py", line 86, in image_warp
w, v = solve_interpolation(dst_pts, flow_pts, device)
File "/root/DST/warp.py", line 226, in solve_interpolation
w_v, _ = torch.solve(rhs, lhs)
RuntimeError: solve_cuda: U(371,371) is zero, singular U.

I tested 10 image pairs, 2 of them got this error.

Update: this seems to be a random event. For the same pair of images, it may or may not happen.

When the image aspect ratio is large, after resizing the image with respect to the long edge, the shorter edge length becomes too small and the short edge will vanish during feature extraction because of the downsampling process.

Optimizing at scale 1, image size (40, 64)
Traceback (most recent call last):
File "main.py", line 115, in
device=device)
File "/root/DST/styletransfer.py", line 113, in DST
feat_e = extractor.forward_samples_hypercolumn(style_im_scaled, samps=1000)
File "/root/DST/vggfeatures.py", line 36, in forward_samples_hypercolumn
feat = self.forward(X)
File "/root/DST/vggfeatures.py", line 32, in forward
feat = self.forward_base(x)
File "/root/DST/vggfeatures.py", line 24, in forward_base
x = self.vgg_layersi
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/module.py", line 722, in _call_impl
result = self.forward(*input, **kwargs)
File "/opt/conda/lib/python3.7/site-packages/torch/nn/modules/pooling.py", line 159, in forward
self.return_indices)
File "/opt/conda/lib/python3.7/site-packages/torch/_jit_internal.py", line 247, in fn
return if_false(*args, **kwargs)
File "/opt/conda/lib/python3.7/site-packages/torch/nn/functional.py", line 576, in _max_pool2d
input, kernel_size, stride, padding, dilation, ceil_mode)
RuntimeError: Given input size: (512x1x4). Calculated output size: (512x0x2). Output size is too small