Can I somehow limit the number of detected humans to 1 during either the training or the inference of the models from trt_pose?

I am using the NVIDIA trt_pose repository for training the model. I then use a code similar to this one for inferencing with the trained model.

My question is: Can I somehow force the models from trt_pose to detect only 1 human, either during training or during inference?

My 2 cents are that since models from trt_pose are trained on the MSCOCO dataset which contains crowds of people on a lot of images, the model is going to be trained to detect multiple people, which will on rare occasions lead to false positives when there is only one person present.

Am I correct? Do you have any suggestions or ideas as to how could I force a model to predict only one person, either during training or inference?

Thank you in advance!

Hi @mjuric1,

We recommend you to raise your concern on trt_pose issues, to get better help.

Thank you.

I will.

In the meantime, I found a potentially useful issue on the aforementioned GitHub repository that is in line with what I want. I’ll open a new issue if this issue doesn’t do what I want.

Thank you for responding!

1 Like