bcmi / bargainnet-image-harmonization Goto Github PK

1. when runing python test.py --name <experiment_name> --model bargainnet --dataset_mode iharmony4 --is_train 0 --norm batch --preprocess resize --gpu_ids 0 --input_nc 20 --netG s2ad
   
   What computer configuration is required to run this project
2. How to deal with my own image, if I have a directly synthesized image, how to harmonize？
   What data is needed to achieve it
   python test.py --name <experiment_name> --model bargainnet --dataset_mode iharmony4 --is_train 0 --norm batch --preprocess resize --gpu_ids 0 --input_nc 20 --netG s2ad
   
   This is the result of my running, it’s not right, I don’t know where the problem is
3. Is it possible to process 4k video streams in real time，That is, what is the processed fps and the maximum resolution of the processed image

While evaluating the model, it seems like the real (target) image has to be passed as input for the discriminator to create a bg_style_map, which is then passed through the generator concatenated with the composite. What should be done when we do not have the real image? Isn't the aim to find the real image?

How to set other batchsize？

hi, this is a great job! I noticed that the output images were all compressed to 256 x 256d size. What should I do if I want to keep the size of the input original?
Looking forward to your reply!

hi, this is a great job! I want to apply the pre-trained model to the animation image domain, but I find that the performance is not as good as that of the real image domain. I hope to get your suggestions on how to improve the performance of the model in animation image domains:

1. Can the model performance be improved by modifying the parameters? What parameters should be changed?
2. If I need to train the model, should I fine-tune the model or retrain the model? What should I do if I want to fine-tune the model?
3. Could you share some information on what setup you used for the training of the model? how many Gpus/for how long?
4. The output image size has been changed to 256*256, how to keep the output image size of the original image?

（a）composite image

（b）Code generated image

Looking forward to your reply!

when i download trained model ，put it under ./checkpoints/<experiment_name>/. <experiment_name> What is this? it is latest_net_E.pth or latest_net_G.pth？
when run test.py : python test.py --name latest_net_G.pth --model bargainnet --dataset_mode iharmony4 --is_train 0 --norm batch --preprocess resize --gpu_ids 0 --input_nc 20 --netG s2ad

i have a error,FileNotFoundError: [Errno 2] No such file or directory: './checkpoints\latest_net_G.pth\test_opt.txt',what is test_opt.txt.i
don,t have this file.

Hello, thanks for opening source the great project.

I have 2 issuss when reproduce the project:

1. What is the exact data structure of iharmony4? Do you train the 4 subclass once? When I run the training, I have bugs for data loading. And I have not change the original iharmony4 data structure.
2. Whear is the code to predict the inharmony level of a composite image by calculating the Euclidean distance between two domain?

thanks in advance!

Can you please update the license of the repo? Thanks!

Hi, I'd like to give this a try but I'm having trouble downloading the models. Is there another location that you could upload them? Thanks very much!

您好，请问一下代码是只支持单卡训练么？可以进行多卡训练么

When I try to run this repo, the error is reported.
"AttributeError: module 'albumentations.augmentations.transforms' has no attribute 'Resize'"
details:
File "/BargainNet-Image-Harmonization/data/base_dataset.py", line 68, in get_transform
transform_list.append(transforms.Resize(256,256))