Segformer classification model(FAN hybrid tiny backbone) evaluation issue

Accelerated Computing Intelligent Video Analytics TAO Toolkit
1

Please provide the following information when requesting support.

• Network Type (Segformer model: FAN hybrid tiny backbone)
• TLT Version : TAO v5.5.0
• Training spec file:
train_spec parameters.txt (1.1 KB)

Hi,

I have trained Segformer model(FAN hybrid tiny backbone) model for classification task on 2 classes (Cat and Dog) with 200 epochs. I have observed a strange behaviors on inference and evaluation stage. The results are saved in the excel sheet and it is observed that all the images is predicted as “cat” category. I have checked with placing of images correctly in folder format and also the class_map.json file in correct order.

Not sure if I have missed anything in train_spec parameters file attached.

Can you please let me know if I can add any advanced parameters in training spec file for getting correct predictions on evaluation stage.

Thanks in advance

3

Could you run classification-pyt notebook successfully? Please see tao_tutorials/notebooks/tao_launcher_starter_kit/classification_pyt/classification.ipynb at main · NVIDIA/tao_tutorials · GitHub . Its training spec file is in tao_tutorials/notebooks/tao_launcher_starter_kit/classification_pyt/specs/train_cats_dogs.yaml at main · NVIDIA/tao_tutorials · GitHub.

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.