Open yitian8377 opened 3 years ago
As I known, there is NO direct way to implement depthwise conv in NVDLA. So, when I use hardware modules to implement depthwise separable convolution , what should I do best?
As I known, there is NO direct way to implement depthwise conv in NVDLA. So, when I use hardware modules to implement depthwise separable convolution , what should I do best?