implementation for the segmentation?

[apple2373]

Hi,

I could not find the implementation for the segmentation and its evaluation.... Where is it?

Best,
Satoshi

[ghost]

@apple2373
this may be helpful for you
https://github.com/lucasb-eyer/pydensecrf

[KeyKy]

I also want to learn the segmentation code~

[Ferenas]

Actually you have to implement your seg masks based on the paper by yourself. After trained from datasets, the net can be directluy used for the mask Inference. The output of the class-wise pooling (attention, the code writer links the class-wise and spatial-wise into one pooling module by Sequentail, so you have to split them in the net to get the class-wise pooling, whose outputs are 2077 in VOC12) , then you choose the maxscore in each channel(classes) to generate the final seg masks

[liuyue718]

Hello, the segmentation mask I got is 1/32 of the size of the input image. To get the segmentation result of the same size as the input image, do I need to directly sample the original size of the image?Isn't that a lot of resolution lost