Are there any ways to accelerate the generating time?

[ThaiNguyen54]

Hi, your work is great. When I use RTX 4070 GPU, it takes about 10 minutes to generate an image, so I wonder if there are any methods that can reduce the generating time to about 10 seconds?

[ArsalanAli915]

@myoons @Taeu please let me know that we can reduce the time to 4 mins?

[prabhasreddy-57]

(hairstyle) C:\Users\prabhas\Desktop\Style-Your-Hair>python main.py --input_dir ./ffhq_image/ --im_path1 source.png --im_path2 target.png --output_dir ./style_your_hair_output/ --warp_loss_with_prev_list delta_w style_hair_slic_large --save_all --version final --flip_check
Loading StyleGAN2 from checkpoint: pretrained_models/ffhq.pt
torch.Size([512])
Setting up Perceptual loss...
Loading model from: C:\Users\prabhas\Desktop\Style-Your-Hair\losses\lpips\weights\v0.1\vgg.pth
...[net-lin [vgg]] initialized
...Done
flip is better, kp_diff : 29.72156524658203 > kp_diff_flip : 19.58245277404785
Number of images: 2
Images: 0%| | 0/2 [00:00<?, ?it/s]source
Images: 50%|█████████████████████████████████████████████████████████████▌ | 1/2 [04:48<04:48, 288.43s/it]target_flip
Images: 100%|███████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [09:37<00:00, 288.89s/it]
Number of images: 2
Images: 100%|████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 2/2 [02:01<00:00, 60.58s/it]
Loading StyleGAN2 from checkpoint: pretrained_models/ffhq.pt
torch.Size([512])
Setting up Perceptual loss...
Loading model from: C:\Users\prabhas\Desktop\Style-Your-Hair\losses\lpips\weights\v0.1\vgg.pth
...[net-lin [vgg]] initialized
...Done
Setting up Perceptual loss...
Loading model from: C:\Users\prabhas\Desktop\Style-Your-Hair\losses\masked_lpips\weights\v0.1\vgg.pth
...[net-lin [vgg]] initialized
...Done
Warp Target Step 1: 0%| | 0/100 [00:00<?, ?it/s]C:\Users\prabhas\anaconda3\envs\hairstyle\lib\site-packages\torch\nn\functional.py:3500: UserWarning: The default behavior for interpolate/upsample with float scale_factor changed in 1.6.0 to align with other frameworks/libraries, and now uses scale_factor directly, instead of relying on the computed output size. If you wish to restore the old behavior, please set recompute_scale_factor=True. See the documentation of nn.Upsample for details.
"The default behavior for interpolate/upsample with float scale_factor changed "
Traceback (most recent call last):
File "main.py", line 162, in
main(args)
File "main.py", line 72, in main
align.align_images(im_path1, im_path2, sign=args.sign, align_more_region=False, smooth=args.smooth)
File "C:\Users\prabhas\Desktop\Style-Your-Hair\models\Alignment.py", line 290, in align_images
save_intermediate=save_intermediate, is_downsampled = is_downsampled)
File "C:\Users\prabhas\Desktop\Style-Your-Hair\models\Alignment.py", line 158, in create_target_segmentation_mask
im2, warped_latent_2, warped_down_seg = self.warp_target(img_path2, src_kp_hm, None, img_path1) # Warping !!
File "C:\Users\prabhas\Desktop\Style-Your-Hair\models\Alignment.py", line 566, in warp_target
latent_in, warped_down_seg = self.optimize_warping(pbar, optimizer_warp_w, latent_W_optimized, latent_F_optimized, mode, is_downsampled, src_kp_hm, im_name_1, im_name_2, cur_check_dir, img_path1, img_path2)
File "C:\Users\prabhas\Desktop\Style-Your-Hair\models\Alignment.py", line 782, in optimize_warping
loss.backward()
File "C:\Users\prabhas\anaconda3\envs\hairstyle\lib\site-packages\torch\tensor.py", line 245, in backward
torch.autograd.backward(self, gradient, retain_graph, create_graph, inputs=inputs)
File "C:\Users\prabhas\anaconda3\envs\hairstyle\lib\site-packages\torch\autograd_init_.py", line 147, in backward
allow_unreachable=True, accumulate_grad=True) # allow_unreachable flag
RuntimeError: no valid convolution algorithms available in CuDNN