Zero-shot cross-dataset generalization
We directly evaluate models that are only trained on ObMan and MOW datasets and report their reconstruction results on HO3D dataset. Both models without finetuning still outperform baselines trained on HO3D dataset. Interestingly, even though the MOW dataset only consists of 350 training images, which is significantly less compared to 21K images from the synthetic dataset, learning from MOW still helps cross-dataset generalization. It indicates the importance of diversity for in-the-wild training.
Input
Ground Truth
Trained on Obman
Trained on MOW
Paper
Bibtex
@inproceedings{ye2022hand,
author = {Ye, Yufei
and Gupta, Abhinav
and Tulsiani, Shubham},
title = {What's in your hands?3D Reconstruction of Generic Objects in Hands},
booktitle = {CVPR},
year={2022}
}