In deepstream pose classifiction, the bias toward walking standing is very severe

Please provide the following information when requesting support.

• Hardware (T4/V100/Xavier/Nano/etc) RTX3090
• Network Type (Detectnet_v2/Faster_rcnn/Yolo_v4/LPRnet/Mask_rcnn/Classification/etc) YoloV8, BodyPose3d, PoseClassification
• TLT Version (Please run “tlt info --verbose” and share “docker_tag” here)
• Training spec file(If have, please share here)
• How to reproduce the issue ? (This is for errors. Please share the command line and the detailed log here.)

I am testing by changing only the video based on the example code here, and the values ​​for standing and walking are very high in the sitting image. Is there any way to solve this?

What type of data sets do you have for learning?

sitting_down, 0.000859, getting_up, 0.001437, sitting, 0.009407, standing, 0.460987, walking, 0.502175, jumping, 0.025136, 
sitting_down, 0.000857, getting_up, 0.001434, sitting, 0.009399, standing, 0.462083, walking, 0.501098, jumping, 0.025130, 
sitting_down, 0.000854, getting_up, 0.001430, sitting, 0.009399, standing, 0.463774, walking, 0.499422, jumping, 0.025121, 
sitting_down, 0.000852, getting_up, 0.001427, sitting, 0.009392, standing, 0.464831, walking, 0.498382, jumping, 0.025117, 
sitting_down, 0.000849, getting_up, 0.001423, sitting, 0.009376, standing, 0.465962, walking, 0.497277, jumping, 0.025112, 
sitting_down, 0.000846, getting_up, 0.001420, sitting, 0.009368, standing, 0.467028, walking, 0.496231, jumping, 0.025107, 
sitting_down, 0.000844, getting_up, 0.001417, sitting, 0.009367, standing, 0.468695, walking, 0.494580, jumping, 0.025097, 
sitting_down, 0.000841, getting_up, 0.001414, sitting, 0.009360, standing, 0.469728, walking, 0.493565, jumping, 0.025092, 
sitting_down, 0.000838, getting_up, 0.001410, sitting, 0.009343, standing, 0.470849, walking, 0.492473, jumping, 0.025088, 

We need some time to verify the accuracy of the model and the sample.

I think it would be good if you could share the results of your progress.

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

Please refer to DeepStream Pose Classification predictions are highly biased towards Walking - Intelligent Video Analytics / DeepStream SDK - NVIDIA Developer Forums

Moving to TAO forum.

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