Segformer Classification model inference issue

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Hi, after converting the segformer’s fan_hybrid_tiny.pth classification model file to onnx format for checking the model inference, I got the low confidence score like less than 1 %. So, i thought to use the exact model’s architecture while converting the pth file to onnx file. but i couldn’t find the exact architecture code. Please let me know where can i find the exact architecture and also any suggestion in improving the accuracy

2

Is there any issue here? For tao issues, you can create a forum topic in TAO forum: TAO Toolkit - NVIDIA Developer Forums. Thanks!

3

Moving to TAO forum for better tracking. @kurapati.venkatesh

4

The code for segformer is in tao_pytorch_backend/nvidia_tao_pytorch/cv/segformer at main · NVIDIA/tao_pytorch_backend · GitHub.