Only detect document edges

[UserJTZhang]

Thank you very much for your algorithm. I am currently using it for document edge detection. In this process, I do not want to detect the edges of document content (text or image content), nor do I want to detect the edges of the background, only keeping the edges of the document. But obviously, the results I obtained contain too many other edges, and I see that the feature maps given by out1 and out2 also contain many other edges. As shown in the figure, I have converged after 24 epochs of training, with approximately 50000 pairs of image data. I hope to receive your reply and provide some directional suggestions. Thank you very much.

[UserJTZhang]

[xavysp]

Thank you very much for your algorithm. I am currently using it for document edge detection. In this process, I do not want to detect the edges of document content (text or image content), nor do I want to detect the edges of the background, only keeping the edges of the document. But obviously, the results I obtained contain too many other edges, and I see that the feature maps given by out1 and out2 also contain many other edges. As shown in the figure, I have converged after 24 epochs of training, with approximately 50000 pairs of image data. I hope to receive your reply and provide some directional suggestions. Thank you very much.

Hi, sorry for my late response: I think you need more samples, you just have 1747 iterations for an epoch. Mine are around 56k per epoch. As this is like a segmentation you should you some preprocessing or post processing in this task.

[KeyWu]

Thank you very much for your algorithm. I am currently using it for document edge detection. In this process, I do not want to detect the edges of document content (text or image content), nor do I want to detect the edges of the background, only keeping the edges of the document. But obviously, the results I obtained contain too many other edges, and I see that the feature maps given by out1 and out2 also contain many other edges. As shown in the figure, I have converged after 24 epochs of training, with approximately 50000 pairs of image data. I hope to receive your reply and provide some directional suggestions. Thank you very much.

您好，想请问您一下我在执行python main.py --choose_test_data=-1时报错，您有时间能回复一下吗？我将不胜感激！