zhengqili / cgintrinsics Goto Github PK

Hello,

First of all, thank you for your wonderful work and deciding to make it publicly available! I am trying to run your training scripts. I have followed the instructions in your Readme and have successfully set up the training pipeline and it is running. But I noticed that even though in the opts it says save freq is 5 epochs, I don't see anything saved under checkpoints. I have not modified anything apart from what was instructed in the readme. I also don't see any saving subsection code in the training script. Could you please confirm this? From the original pix2pix code, it would need a model.save_networks function call. Thank you.

Hi. As I'm reading your awesome project code, I got wondering part.
I think gradient deleting code needs to be located before forward()
But now, gradient deleting code is being located after forward()
In other words, calculated gradients are being deleted every time

I think this part
https://github.com/lixx2938/CGIntrinsics/blob/master/models/intrinsic_model.py#L99
needs to be fixed like:

self.optimizer_G.zero_grad()  
self.forward_both()

Please let me hear your opinion whether my fixing is correct or not.

Dear lixx2938,

I read your paper <CGIntrinsics: Better Intrinsic Image Decomposition through Physically-Based Rendering>, it is a nice paper, I really like it.

By use the SunCG 3D models and CGIntrinsics rendering methods, could I remove one object from the 3D model and re-rendering the image? We want to create images pairs that a specific object exists and doesn't exist.

In your paper, you mention that it takes 30 mins to rendering one image. So it may take a long time to re-rendering the image, right? And do you publish the rendering code? or just the rendered images? Sorry, I could not find the rendering code.

Thank you very much for your help.

Best Wishes,

Alex

Hi,
thank you for your work, I like the results. We are thinking about utilizing the model to improve the registration pipeline for pairs of images. However, I am not successful in using the code for testing on our dataset. I would find it very useful if you could provide some info about how to get predictions from inputs which are not in the IIW/SAW dataset format - e.g. how to get an output from a single image.

Best regards,
Milan Pultar

I first got the "[Errno 32] Broken pipe" error, then I changed the number of workers in the DataLoader to 0, and it solved the problem. Now, when I run train.py, I get the Runtime error below:

Traceback (most recent call last):

File "", line 1, in
runfile('E:/Research/img_decomp/CGIntrinsics/train.py', wdir='E:/Research/img_decomp/CGIntrinsics')

File "E:\Anaconda\lib\site-packages\spyder\utils\site\sitecustomize.py", line 710, in runfile
execfile(filename, namespace)

File "E:\Anaconda\lib\site-packages\spyder\utils\site\sitecustomize.py", line 101, in execfile
exec(compile(f.read(), filename, 'exec'), namespace)

File "E:/Research/img_decomp/CGIntrinsics/train.py", line 105, in
model.optimize_intrinsics(epoch, data_set_name)

File "E:\Research\img_decomp\CGIntrinsics\models\intrinsic_model.py", line 98, in optimize_intrinsics
self.forward_both()

File "E:\Research\img_decomp\CGIntrinsics\models\intrinsic_model.py", line 81, in forward_both
self.prediction_R, self.prediction_S = self.netG.forward(self.input_images)

File "E:\Research\img_decomp\CGIntrinsics\models\networks.py", line 1819, in forward
return nn.parallel.data_parallel(self.model, input, self.gpu_ids)

File "E:\Anaconda\lib\site-packages\torch\nn\parallel\data_parallel.py", line 183, in data_parallel
inputs, module_kwargs = scatter_kwargs(inputs, module_kwargs, device_ids, dim)

File "E:\Anaconda\lib\site-packages\torch\nn\parallel\scatter_gather.py", line 35, in scatter_kwargs
inputs = scatter(inputs, target_gpus, dim) if inputs else []

File "E:\Anaconda\lib\site-packages\torch\nn\parallel\scatter_gather.py", line 28, in scatter
return scatter_map(inputs)

File "E:\Anaconda\lib\site-packages\torch\nn\parallel\scatter_gather.py", line 15, in scatter_map
return list(zip(*map(scatter_map, obj)))

File "E:\Anaconda\lib\site-packages\torch\nn\parallel\scatter_gather.py", line 13, in scatter_map
return Scatter.apply(target_gpus, None, dim, obj)

File "E:\Anaconda\lib\site-packages\torch\nn\parallel_functions.py", line 89, in forward
outputs = comm.scatter(input, target_gpus, chunk_sizes, ctx.dim, streams)

File "E:\Anaconda\lib\site-packages\torch\cuda\comm.py", line 148, in scatter
return tuple(torch._C._scatter(tensor, devices, chunk_sizes, dim, streams))

RuntimeError: CUDA error: invalid device ordinal (exchangeDevice at C:/a/w/1/s/tmp_conda_3.6_090826/conda/conda-bld/pytorch_1550394668685/work/aten/src\ATen/cuda/detail/CUDAGuardImpl.h:28)
(no backtrace available)

I did not change the --gpu_ids. What number should I change it to?

Thanks for your great project. However, I have trouble visualizing the predictions. The image looks very strange, could you please tell me how to correctly visualize them?

I ran the code and visualized the results, but I got a
dull image that was very different from the paper.

How do I get the same results as the paper?

Hello! Thanks for sharing your work. I wonder if the final model is fine-tuned on the MIT dataset, or only trained on CGI, IIW anf SAW. Could you please answer that? Thank you