Can you please provide the configuration files for your ablation studies? Particularly:
Training POEMv1, MvP, PE-MeshTR, FTL-MeshTR on Sharded Datasets [This repo uses unsharded data]
Training POEMv2 on each dataset separately: I can probably do this by updating provided released config such that "MIX_RATIO: 1" and "DATASET_LIST = ["DexYCB"], Right?
The ablation studies in question are conducted using POEM-v1 (i.e. the original version), so unfortunately we haven't adapted POEM-v2 for these configurations.
Yes, that is correct. It's advised to train the model using similar hyper-parameter settings as POEM-v1
Hello,
Can you please provide the configuration files for your ablation studies? Particularly:
Thanks.