I changed myself Yolov5s model, so the architecture is practically the same as you have. It works fine on cpu, but when I’m using the quantize tool and run yolov5s_timvx it gives me the error (input size 86400 != reshaped(345600). I’m using my custom model of pictures 640x640 with 13 classes and I cann’t understand what’s happening.
I’ll repeat that Cpu version works fine, the problem is in quantize model. So what is with the quantize tool wrong.
And please can you finally answer? I’ve been waiting a lot in different topics, which still not answered.
Here is my github, where the python model I use (I changed yolo.py and export.py)