search

Joker316701882 / Deep-Image-Matting

This is tensorflow implementation for paper "Deep Image Matting"
626 stars 188 forks source link

crop problem & evaluation codes #5

Open mikirui opened 6 years ago

mikirui commented 6 years ago

Hi,

I read your Chinese blog, and I think maybe you don't need to padding for 'crop security'. I did it by specify the region I can crop before I randomly crop centered on the unknown region pixel, but I implement it with caffe. By the way, I think the way to compute evaluation metrics like 'Gradient' and 'Connectivity' are of ambiguity, do you have codes for that? It would be great help to me. Thanks~

rui

hector-sab commented 6 years ago

This might be off-topic but could you share the link of the blog?

Héctor

Joker316701882 commented 6 years ago

@hector-sab Sorry it's in Chinese http://calebml.leanote.com/post/Deep-Image-Matting%E5%A4%8D%E7%8E%B0%E8%BF%87%E7%A8%8B%E6%80%BB%E7%BB%93

Joker316701882 commented 6 years ago

@mikirui Thank you for your great advice! But I don't quite understand your question. Gradient is automatically calculated by dl framework. If you mean 'loss structure', it can be seen in 'readme'.

mikirui commented 6 years ago

@Joker316701882 Sorry, I don't mean that gradient, actually I mean the metrics to measure the performance of Matting methods: 'SAD', 'MSE', 'Gradient' and 'Connectivity', which is presented in Table 1 of this paper. So maybe you don't focus on this, thank you anyway~~