cm-jsw / graphcut Goto Github PK

在构建graph的时候，所有的节点跟S/T点相连的边，如果不是种子点，那么应该是构建灰度直方图，这样来给定权重的吧？为什么代码中全给0了？

                '''d_f = np.power(self.image[y, x] - self.foreground_average, 2)
                d_b = np.power(self.image[y, x] - self.background_average, 2)
                d_f = np.sum(d_f)
                d_b = np.sum(d_b)
                e_f = d_f / (d_f + d_b)
                e_b = d_b / (d_f + d_b)'''
                self.nodes.append((self.get_node_num(x, y, self.image.shape), 0, 0))

这个真的可以运行出结果吗，为什么总是connect失败

Hello, thanks of your contribution about such good work.
I am using graph neural network to parcel brain region, however the result is not good, I want to use graph cut as post process.
The inputs of my current model are: adjacency matrix (10242, 10242), feature matrix (10242, 6), label matrix (10242). The output is the probability y that each node belongs to a label, and its dimension is (10242, 36)
I want to use graph cut to update y for better performance.

I have a few questions about your code:
First, most of graph cut is only for two categories. Can you do multi label tasks?
Second, if I want to do post-processing, what should my input be?