search

minar09 / ACGPN

"Towards Photo-Realistic Virtual Try-On by Adaptively Generating↔Preserving Image Content"，CVPR 2020. (Modified from original with fixes for inference)
https://github.com/switchablenorms/DeepFashion_Try_On
77 stars 70 forks source link

RuntimeError: CUDA error: device-side assert triggered #14

Open shashini4296 opened 3 years ago

shashini4296 commented 3 years ago

Traceback (most recent call last): File "E:\Virtual try on\ACGPN-master\test.py", line 122, in main() File "E:\Virtual try on\ACGPN-master\test.py", line 97, in main fake_image, warped_cloth, refined_cloth = model(Variable(data['label'].cuda()), Variable(data['edge'].cuda()), Variable(img_fore.cuda()), Variable( File "C:\Users\SM\AppData\Roaming\Python\Python39\site-packages\torch\nn\modules\module.py", line 889, in _call_impl result = self.forward(*input, *kwargs) File "E:\Virtual try on\ACGPN-master\models\pix2pixHD_model.py", line 492, in forward return self.inference(label, pre_clothes_mask, img_fore, clothes_mask, clothes, all_clothes_label, real_image, pose, grid, mask_fore) File "E:\Virtual try on\ACGPN-master\models\pix2pixHD_model.py", line 384, in inference input_label, masked_label, all_clothes_label = self.encode_input( File "E:\Virtual try on\ACGPN-master\models\pix2pixHD_model.py", line 233, in encode_input 1, (label_map (1 - clothes_mask)).data.long().cuda(), 1.0) File "C:\Users\SM\AppData\Roaming\Python\Python39\site-packages\torch\tensor.py", line 528, in rsub return _C._VariableFunctions.rsub(self, other) RuntimeError: CUDA error: device-side assert triggered

How can I fix this issue?

minar09 commented 3 years ago

It might be related to your work environment, I think you may find solutions online. For instance: https://stackoverflow.com/questions/51691563/cuda-runtime-error-59-device-side-assert-triggered

shashini4296 commented 3 years ago

@minar09 Thank you very much. I found the issue is with the number of labels. I'm using PGN (https://github.com/Engineering-Course/CIHP_PGN.git) to create the labels. How can I fix this issue or is here any other better way to create labels.

minar09 commented 3 years ago

Can you please clarify more details on your issue? Do you mean you have a mismatch with the number of labels or something else? CIHP_PGN generated segmentation labels number should be the same here in ACGPN, which means you need to update the code accordingly.

shashini4296 commented 3 years ago

I need to get those 14 segmentation labels from the CHIP_PGN. Can you tell me how to update the code.

Background Hair Upclothes Left-shoe Right-shoe Noise Pants Left_leg Right_leg Left_arm Face Right_arm

@minar09 Thank you very much.

minar09 commented 3 years ago

Hi @shashini4296 , if your dataset images contain new segments that are not present in CIHP_PGN, and you need those to be labeled separately, then you need to train the network with your custom dataset and annotations. However, if your desired image labels are somewhat similar to CIHP_PGN format, you can generate the segmentation with the pretrained CIHP_PGN model first, and then assign new label values to the segmentation according to your needs. Hope that helps. Thanks!

shashini4296 commented 3 years ago

@minar09 Thank you very much for the help