About training using Obman

[powenlo]

Thanks for your great work. I have several questions about training using Obman.

I have just run python traineval.py --atlas_predict_trans --atlas_predict_scale --atlas_mesh --mano_use_shape --mano_use_pca --freeze_batchnorm --atlas_separate_encoder. After training for 200 epochs, the object reconstruction is still not as good as the demo you provided. I am wondering to know whether it's because the number of epochs is not enough or I need to do extra fine-tuning. If yes, could you provide me with some guidance on fine-tuning please.

Thanks a lot.