Training st_movenet_thunder_heatmaps

[olofhq]

Hi,

I've started trying to train a keypoint detection model using the "training" flag in the config file. I started of with the lightning-model but did not get the performance I need. I tried to instead use the thunder model but I found that in "stm32ai-modelzoo-services/pose_estimation/src/utils/models_mgt.py" there is no case for the thunder model. Only "st_movenet_lightning_heatmaps" and "custom". Is there a way around this?

I'm using a completely custom dataset of a different object with 7 keypoints. This also lead to me having to add a "f4()" function in "stm32ai-modelzoo-services/pose_estimation/src/evaluation/metrics.py"

Thanks for help.