jetson-inference
jetson-inference copied to clipboard
Models with NHWC format
Hi @ligaz,
Thank you for the great work!
I have an ONNX model with its input and output nodes shaped in NHWC format as it was exported from TensorFlow. I used some tools to modify it to NCHW format but the output were corrupted (meaningless). I also have another model with its input nodes in NCHW format and its output nodes in NHW (I also changed the DIMS_H
and DIMS_W
and tried). This also gives meaningless result.
I was wondering if you know how I can work with such models to get a correct result? or if jetson-inference
supports NHWC format?
Thank you so much!