Hi, Tang. I found another bug which is similar to #24: the args passed get_x_2d ought to be (w, h). Unlike #24, it didn't raise an explicit error when height: width of inputs is not 1:1, this part of code just returns wrong correspondence in that case, which made my training failed.
Hi, Tang. I found another bug which is similar to #24: the args passed get_x_2d ought to be (w, h). Unlike #24, it didn't raise an explicit error when height: width of inputs is not 1:1, this part of code just returns wrong correspondence in that case, which made my training failed.
Thank you.