Test in my own dataset got 'blue' result.

[JustinhoCHN]

Hello @nothinglo , I've trained the model with unsupervised hdr dataset for about 100 epochs, if I use the test image from fivek dataset, I got pretty good result: (GT/Predicted)

But when I test it with my own dataset, the result looks like this:

Some regions in those test images are changed to blue, I suspected that there're so many blue sky in the training dataset (both in input and label_HDR).

And I tested these images in the demo website, they look like this:

Here're ground truth images:

The next step I want to improve the result will be adding more own dataset to finetune the model. Anyone has any idea about why it become so blue and how to solve it? Thanks in advanced.

By the way, here's my loss curve: #40

[liulizhou]

have you solved the problem ?

[JustinhoCHN]

@liulizhou nope, you are the first reply.... maybe it's a dataset domain missmatch problem.

[liulizhou]

@JustinhoCHN I also train the program, now 130 epoch. Next i will test it and maybe i will have the same problem.

[liulizhou]

@JustinhoCHN now i have not test it, but your result may be currect. I think you should check the ouput channel, if the output is RGB, you should change it to BGR and then show it with opencv (the result is like this)... maybe it's not.

[JustinhoCHN]

@liulizhou if you done training, you can use my image to test it, one of my friend's result like this:

[xuanzhangyang]

@JustinhoCHN Yes, it‘s my result^v^...

[liulizhou]

Hi @JustinhoCHN @xuanzhangyang i test two own images and the result is like this. the first result and ground truth. the second result and ground truth.

[JustinhoCHN]

@liulizhou seems faces turn to yellow, white background or sky turn to blue, and bringing artifacts.

[liulizhou]

@JustinhoCHN Yes, we should remove the artifacts. i have no idea why this happen. I test DPED and HDRNET, and the results also bring artifacts.

[ExplorerLGD]

我发现在物体的边缘位置，会出现一些模糊，很影响效果

[xiaozhi2015]

Here is my result.

I'd like to ask, do you know how to expand the inference dimension to 2048? @JustinhoCHN @liulizhou I'll appreciate it very much for your answers.

[nothinglo]

@JustinhoCHN I do think it is dataset domain mismatch problem. @xiaozhi2015 I just uploaded the inference models and the code I used in my demo website.

[xiaozhi2015]

Much thx for your great work! @nothinglo

[TriLoo]

Have you resolved this problem? i.e., face turned to yellow, and white background turned to blue ? @liulizhou