qingsenyangit/ahdrnet

Attention-guided Network for Ghost-free High Dynamic Range Imaging

ahdrnet's Issues

exposure time

hi, thanks for your work.

as mentioned in your paper, the training HDR (LDR^2.2/exposure_t) is generated.
But I check the dataset from Kalantari there is 0.0 value in exposure_t. What did you do to deal with that?

Thanks!

When can you upload the training code？

I tried to reproduce your training code, but it didn't work. When can you upload the training code？

some functions for test are missing

Could not find testing_fun() and testimage_dataloader() in the whole project when running Script_testing.py.

will you release the training code and the dataset? looking forward~

I tried to test your model but don't seems to work

Can we have the script for data preparation? Probably is because you subtracted the mean and dived for std. Let me know :)

正在尝试训练你的网络，能告诉下训练时输入数据的rgb值压缩到了多大吗

Can you upload the dataset or ./matlab_script? thanks a lot.

Can you give more train details or release train code

Hi. Thanks for your great work!
I realized the train code based on your code, set lr=1e-5, batch_size=8, train 100 epochs, but the PSNR is worse than that in your paper. Can you give more train details or release train code. Thanks!

Using the given model，PSNR-u is lower than the cvpr repoerted

def range_compressor_tensor(x):
const_1 = torch.from_numpy(np.array(1.0)).cuda()
const_5000 = torch.from_numpy(np.array(5000.0)).cuda()
return (torch.log(const_1 + const_5000 * x)) / torch.log(const_1 + const_5000)
Is this fuction not correct?
Thanks

Can you please publish the ./matlab_script？

Where can I get the ./matlab_script？It's hard to test without it.

done

Can you upload the dataset or ./matlab_script and train code? thanks a lot.

Washed-out colors

Hello,
I can't seem to get the kind of beautiful saturated colors from the paper using this code.. I understand that the main contribution of your paper is a ghost-free image, and dealing with motion, not colors. Does this mean you used some additional color mapping or maybe tweaked the colors manually using some image editing software just for the paper? Or am I missing something obvious?

CUDA out of memory

Have you ever met the problem that CUDA out of memory? I have two gpus and both is Tesla T4 16g, even I set the batchsize to 1 ,the problem still happens. Follow is the error code:
Traceback (most recent call last):
File "script_testing.py", line 63, in
loss = testing_fun(model, testimage_dataset, args)
File "/root/AHDRNet/running_func.py", line 185, in testing_fun
output = model.forward(data1, data2, data3)
File "/root/miniconda2/envs/hdr/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 152, in forward
outputs = self.parallel_apply(replicas, inputs, kwargs)
File "/root/miniconda2/envs/hdr/lib/python2.7/site-packages/torch/nn/parallel/data_parallel.py", line 162, in parallel_apply
return parallel_apply(replicas, inputs, kwargs, self.device_ids[:len(replicas)])
File "/root/miniconda2/envs/hdr/lib/python2.7/site-packages/torch/nn/parallel/parallel_apply.py", line 83, in parallel_apply
raise output
RuntimeError: CUDA out of memory. Tried to allocate 2.86 GiB (GPU 0; 14.76 GiB total capacity; 13.19 GiB already allocated; 873.75 MiB free; 11.17 MiB cached)
Thank you

can you please upload training code and the code for your job published in IJCV

I noticed that the following work called DAHDRNet has been published in IJCV. Both AHDRNet and DAHDRNet are excellent works in HDR deghosting. Please release the training code for followers of your nice job.

how to show the attention maps

i want to dump the attention maps just showed in the paper.
in paper, it looks like 3-channels feature
how to convert the 64-channels attention maps to 3-channels maps?
just sum it up?

done

done

How to get qualitative results in this paper ?

Hi, I have some questions about the qualitative results(Figure 7 and Figure 8) in your paper.

Were the tone-mapping results in Figure 7 and Figure 8 obtained using Photomatix Pro? If so, can you share your preset(an XMP file) for your tone-mapping scheme?

Thank you very much!

training code

Can you tell me how many times the training code has been trained