The number of people and their locations are not accurately predicted

[XiaozhenQiao]

Thank you very much for the author's excellent work. We tried to replicate the multi-HMR training process and overfit on a single image. We found that we could correctly predict the pose of people, but not the number of people or their positions. We tried for a long time but couldn't solve this problem. Have you encountered similar problems during training? Do you have any good suggestions? Thank you again and look forward to your reply.