        Docs/Applications/YOLOv10n INT8 Inference on GPU and NPU
      

YOLOv10n INT8 Inference on GPU and NPU

Application DetectionYOLOv10nINT8GPU / NPU

The demo GIF may take some time to load. If it does not appear immediately, please wait.

Inference with the YOLOv10n model was conducted on both NVIDIA AGX Orin and Qualcomm EXMP-Q911 platforms, utilizing their respective GPU and NPU. A confidence threshold of 0.5 was applied, and the visual outputs were found to be highly comparable.

Jetson AGX Orin

Platform information

RAM: 32GB
TensorRT SDK Version: 8.5.2.2

How to Use

We are using the Ultralytics framework as a reference

Convert model:

          {{ cbConvert.fname }}
          
          bash

Inference:

          {{ cbInfer.fname }}
          
          bash

EXMP-Q911

Supported Versions

Choose the iQ-Studio version that matches your platform. To check your BSP version, run cat /etc/innodisk/BSP-version on the platform.

Your platform	iQ-Studio version
BSP v2.5.0 or later	`v0.0.11` or later
BSP earlier than v2.5.0	`v0.0.10`
Ubuntu image `x09`	`v0.0.10`

For v0.0.10, run git checkout v0.0.10 in the iQ-Studio directory before ./install.sh. To return to the latest version, run git checkout main.

Platform information

RAM: 36GB
Qnn SDK Version: 2.38

How to Use

Use the iqs-launcher to start the application.

          {{ cbLaunch.fname }}
          
          bash

If you want to change the video, please put your video in the current directory.

          {{ cbVideo.fname }}
          
          bash

Output location: The predicted video will be saved in the output folder.

Conclusion

In this case, GPU and NPU performance is almost identical, but the NPU has a lower cost and lower power consumption.