junjun-jiang / bambnet Goto Github PK

您好，
如果我没有理解错误的话，在您的文章中，是用defocusmask将一张图图像分成4张图像分别进入预设好的sub-restorer,但是在您的模型包括训练和测试文件当中这个好像没有体现出来，不知道是我的理解错还是少看了文章的部分，能不能请您解答一下呢？

Thanks for providing the code for the BaMBNet, I think it’s a very interesting method!
However , it only can be train by single GPU, it will take a lot of time to reproduce the performance.
I just want to see the results generated by the model, can you provide your pretrained model ?
Thanks again for your help!

How to generate the COC map, and Where is the COC map codes location?

Many thanks for your work. I am confused about the input for model in test.py. It seems that the COC maps are not used as input for model in test.py. While the COC maps ARE used as input for model in train.py. Is it used for ablation study or a wrong setting. Many thanks.

1. In your code train.py, the model input is (l_img, r_img, b_img). Is b_img COC map of train dataset? But b_img is not obtained in your blur_train.py. So do I need to change your code blur_test.py to obtain b_img?
2. In your code test.py, the model input is (l_img, r_img, r_img).If there’s been a mistake？Why do you concatnate two r_img?

Hi Guys,

Want to read your paper but seems the arXiv link cannot be opened by me.
Could you check it or send it to me offline ([email protected]) ? Thanks.

In Section III C of your paper, the thresholds r2-4 are updated by adaptively learning from a small amountof meta-data. However, I don't find this part in your code. I only find you give four static_kernel_size . ”Combined with the model training process, generatingthe defocus mask is a nested optimization“ Where is the nested optimization?

Thanks for projects again!
I used the config you given to train 20 epochs Cocmap on dataset of Canon by the preprocessing code 'image_to_patch_filter.py' , and the inference results are shown in the figure below Figure 1, all pixels is black with the value >1000.There are several questions:
1.The pixel value of the image is uint16, and the effect of the sub-area described in the paper cannot be seen. I would like to know whether my training is wrong or yours is the same. 2.Why three channel and not a single channel, and what's the meaning of the pixel value on each result graph?
In addition, I modified the config "niter: 500000 epoch: 300 #20" , trained it at about 45 epoch(70 000 iter), and the phenomenon of unsupervised loss=0 would appear. Then, the test results were shown in Figure 2, is that normal?