Yu Zheng

感谢你的开源，从mindspore版本作者得知实验结果为pytorch版本训练所得，想用你的代码跑一个对比实验，十分冒昧的的提出这个请求，希望你能同意

Hi, Thank you for sharing your nice work. I also want to know the training data, i.e., which 5000 images are chosen for each training set. Could you share the...

RuntimeError: Given input size: (8x1x1). Calculated output size: (8x0x0). Output size is too small

1

在运行main.py的时候报了这个错，用的是Haze4K数据集，看了下代码发现应该是common.py文件中的第277行class Down有问题，因为atmosphere map为size为(8,1,1,1)，而代码里写了Maxpool2d(2)所以报错，我现在自己这里改了下对于atmosphere map设置了个检测，如果是atmosphere map就Maxpool2d(1)，想问下作者你跑通的代码是怎么写的，我这么写会影响实验么

关于DAD数据集的设置

3

` elif is_jpg_file(input_path):` `AB = cv2.imread(input_path).astype("float") / 255.0` `haze_image, GT = get_patch_da(AB, self.size)` `trans = np.zeros((self.size, self.size, 3))` 这段代码应该是DAD的6000张数据集的代码，DAD应该是有transmission map的，然后没有atmospheric map。但程序里好像是对transmission map和atmospheric map设置为0，但这样与论文里关于3.2 Supervised Loss on Labeled data 有误差，而且这样计算transmission loss...

Question about other network performance.

1

Thank you for your work! Your work has inspired me a lot. I have a little question. In table.3 of your paper, you quantitative compare different methods. But as far...