I train a tensorflow model, and complie to tvm, but after do some benchmark test, i find the tvm model is slower than tensorflow, especially tensorflow can inference with bach size > 1. so i try to use atuo tvm to tune my tensorflow model. But the atuo TVM only support NCHW layout and tensorflow’s layout is NHWC. when my input shape is (N, H, W ,C) it will cause error. when change my input shape to (N, C, H, W) , it works. i wonder the change is right? dose tvm do something to transform tensorflow NHWC layout to NCHW automatic? i don’t need to anything, and just reshape my input to NCHW?
