search

feihuzhang / GANet

GA-Net: Guided Aggregation Net for End-to-end Stereo Matching
MIT License
554 stars 135 forks source link

Asking about Guidance class. #54

Open nakeeun opened 5 years ago

nakeeun commented 5 years ago

Thank you for your amazing work. GANet is really impressive with two cost aggregation layer based on SGM, CVF. I have question about your GANet code. When i research your paper and your code, i think code and paper in Eq.(4), Eq.(7) not matched, especially weights. First of all, in module.GANet.py, g0, g1, g2, g3 are weights in SgaFunction. However, in sga_cuda_forward from src.GANet_cuda.cu, those(g0,g1,g2,g3) variable name are changed with guidance_down, up, right, left, respectively. Even these variables go into each sga direction function(such as sga_left_foward, sga_right_backward...) Example is as below. example Do i miss something??

feihuzhang commented 5 years ago

Yes, in the implementation g0 ... g1 means the guidance weights for aggregation in different directions. For examples, g0 means aggregation down which has a set of weight g0={w0, w1... w4}.