mukulkhanna / fhdr Goto Github PK

Would you like to share your code of LDR generation?

start_epoch = 1
model.apply(weights_init)

m.weight.data.normal_(0.0, 0.0)
I take error this code in train.py file=> RuntimeError: normal_ expects std > 0.0, but found std=0

Can you tell me, how can i get

Hi,
Dataset links are either corrupted or not working.
Please rectify this.
Sample test datasets can be downloaded here -

512x512 size images
256x256 size images

Thanks

Hello!
According to data_loader.py, the GT(HDR) image is in range [0, +inf) and normalized the GT by -0.5 and then div 0.5, that is in range[-1, +inf)
the input(LDR) image is in range [0, 255] and also normalized by mean(0.5) and std(0.5) and that is in range[-1, 1]
Your FHDR model has one Tanh layer at the end of the model, so how do you generate output >1 ?

Hi, thank you for your contributions to HDRI area, I really like the job you have done here.
I want to ask if there is an available dataset for training to train model towards more iterations. I'd be happy to see the results.
Looking forward your answer!

Can you please help me get it working?

File "C:\Users\MyPC\Downloads\FHDR-master\FHDR-master\test.py", line 68, in
for batch, data in enumerate(tqdm(data_loader, desc="Testing %")):
File "C:\Users\MyPC\AppData\Local\Programs\Python\Python311\Lib\site-packages\tqdm\std.py", line 1178, in iter
for obj in iterable:
File "C:\Users\MyPC\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\utils\data\dataloader.py", line 633, in next
data = self._next_data()
^^^^^^^^^^^^^^^^^
File "C:\Users\MyPC\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\utils\data\dataloader.py", line 677, in _next_data
data = self._dataset_fetcher.fetch(index) # may raise StopIteration
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\MyPC\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\utils\data_utils\fetch.py", line 51, in fetch
data = [self.dataset[idx] for idx in possibly_batched_index]
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
File "C:\Users\MyPC\AppData\Local\Programs\Python\Python311\Lib\site-packages\torch\utils\data_utils\fetch.py", line 51, in
data = [self.dataset[idx] for idx in possibly_batched_index]
~~~~~~~~~~~~^^^^^
File "C:\Users\MyPC\Downloads\FHDR-master\FHDR-master\data_loader.py", line 52, in getitem
self.hdr_data_path, self.hdr_image_names[index]
~~~~~~~~~~~~~~~~~~~~^^^^^^^
IndexError: list index out of range

Hi, @mukulkhanna

I run a test using the provided 2-Iterations model and the 256x256 test set. The result can be seen as follow:

I notice that the avg PSNR and SSIM are not very high. Am I using code in a wrong way? Or it is because the provided pretrain mode is not optimal?

Hi,LDR data is of type UNIT8 and normalized to [-1,1]，HDR data is of type Float32,if you follow the code, HDR data is not normalized to [-1,1],the value of hdr_tensor is greater than 1 ,the loss value will occur nan.The HDR data processing code is as follows:

transform_list = [
transforms.Lambda(lambda img: torch.from_numpy(img.transpose((2, 0, 1)))),
transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5)),
]
transform_hdr = transforms.Compose(transform_list)
hdr_tensor = transform_hdr(new_hdr)

How can WE effectively normalize HDR data value to [-1,1]?

The paper mentions the HDR-VDP Qscore, but the test file does not contain this evaluation metric. Could you provide this testing metric? Thanks.