Open penguinflux opened 1 month ago
Hi @penguinflux thanks for raising this issue, looks like an issue with FiftyOne when there are no ground truth labels on a frame. I will look into this soon.
@penguinflux apologies for the delay - I'm looking at this now and actually I don't get any errors even if every single ground truth label is NaN. Is it possible that you were labeling frames while this model was training? In that case the ground truth labels would have been updated but the final predictions (computed on the dataset created at the beginning of training) would not be.
@penguinflux checking in on this, did you see my above message?
Hi, every time I run training via Pose-app GUI, it crashes at the last moment before finishing, like the figure attached. The trained model will be kept and could be used to predict new videos. However, I do not know its impact on models trained. Would you give me any suggestions about this? Thank you!