Deepstream-gaze-app: Results are not good

Accelerated Computing Intelligent Video Analytics DeepStream SDK
, , , ,
1

Please provide complete information as applicable to your setup.

**• Hardware Platform (Jetson / GPU)**GPU
• DeepStream Version7.1
• JetPack Version (valid for Jetson only)
• TensorRT Version
**• NVIDIA GPU Driver Version (valid for GPU only)**12.4
• Issue Type( questions, new requirements, bugs)

i have successfully run the deepstream app, but im getting the incorrect output as you can see that the gaze tracker is not accurate at all which can be seen clearly in the below video as well.


image
image1577×841 65 KB

image
image1517×814 64.1 KB

kindly help me here, where can be the issue as i have spend so much time to set it up and now its not working accurately.

3

gaze sample is not supported by DeepStream 7.1

4

sorry it was deepstream 6.1

5

kindly tell me why these results are not good?

6

OK. The DeepStream gaze sample is the demonstration of how to integrate the Gaze Estimation | NVIDIA NGC model, Facial Landmarks Estimation | NVIDIA NGC model and FaceDetect | NVIDIA NGC model with DeepStream. These models are just pre-trained models. They are not product level. You may consult TAO forum for how to improve the pre-trained models’ accuracy. Latest Intelligent Video Analytics/TAO Toolkit topics - NVIDIA Developer Forums

7

what if use any other model from other source whose results are good than others and convert it into onnx and replace it with the gaze model, because the other two models are working fine the only issue os with the gaze model as it is not tracking the eye movement accurately, so what do you think, would it work? for reference i might be using the model from google mediapipe, IRIS tracking model, can you give me your thoughts about it, or any enhance recommendations?

8

The pipeline can be your reference. For the 3rd party model, we have no comments.

9

can you recommend any solutions for this problem, like what should i do now?

10

The pipeline can be your reference. You may consult TAO forum for how to improve the pre-trained models’ accuracy.

11

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

12

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