Unable to train YOLOv6 model with TAO

Accelerated Computing Intelligent Video Analytics TAO Toolkit
,
1

I am currently working on the NVIDIA Jetson Orin Nano Developer Kit and building a real-time safety monitoring system using YOLOv6 for person detection. I also plan to implement transfer learning to differentiate between people wearing safety suits and those not wearing them.

Before proceeding further, I would like clarification on whether YOLOv6 is officially supported or compatible with the NVIDIA TAO Toolkit. I am considering using TAO for transfer learning and optimization, but I am not sure if YOLOv6 can be trained or fine-tuned within the TAO framework.
If not, what would the recommended approach be to train the YOLOv6 model in this case?
Also, any github repository that i can refer to?

5

YOLOv6 is not officially supported in TAO. Suggest you to use RT-DETR network to run training.

Doc: https://docs.nvidia.com/tao/tao-toolkit/6.25.11/text/cv_finetuning/pytorch/object_detection/rt_detr.html

Spec: https://docs.nvidia.com/tao/tao-toolkit/text/cv_finetuning/pytorch/object_detection/rt_detr.html#creating-an-experiment-spec-file

Notebook: https://github.com/NVIDIA/tao_tutorials/blob/main/notebooks/tao_api_starter_kit/tutorials/rtdetr_detection_distillation.ipynb

You can use latest docker(6.25.11) to run training.
Reference:

image
image2358×1159 86.4 KB

More info can be found in slides.

6

Okay, thank you the issue is resolved now

7

can i train my yolov8 model?
so instead of yolov6 i plan to upgrade to yolov8…

8

Hi , YOLOv8 is also not officially supported in TAO. Suggest you to use RT-DETR network to run training.

9

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.