Closed MetaDev closed 5 years ago
I did not combine these two datasets, but it is a valuable attempt to boost estimation accuracy further. I think you can firstly train the network only with the common joints and then finetune the final FC layer to any joint format you want.
Thanks for the advice! When you talk about the final FC layer you're suggesting when a model is trained to do regression directly on the joint positions not heatmaps? For heatmaps it would mean finetuning the last conv layer, correct?
Yes, I use regression in this project.
Have you tried combining both dataset LSP and MPII? LSP has the joint "neck" and MSPII "upper_neck" and "thorax". from the images it seems neck and upper_neck are semantically the same but I'm not sure. Do you have any insight in this?