Views
No views yet
PoseOnly ablation variant of EquiDexFlow.hf_hub_download and torch.load (see snippet below).1from huggingface_hub import hf_hub_download
2import torch
3from equidexflow.models import get_dex_model
4
5model = get_dex_model(hand="allegro")
6path = hf_hub_download("cenwerem/equidexflow-allegro-pose-only", "checkpoint_best.pt")
7ckpt = torch.load(path, map_location="cpu", weights_only=False)
8model.load_state_dict(ckpt["model"])
9# ckpt also carries: "epoch", "optimizer", "best_val_loss"1@article{enwerem_equidexflow_2026,
2 author = {Enwerem, Clinton and Baras, John S. and Belta, Calin},
3 title = {{EquiDexFlow}: Contact-Grounded {SE}(3)-Equivariant Dexterous Grasp Generative Flows},
4 year = {2026},
5 doi = {10.48550/arXiv.2606.12728},
6 number = {{arXiv}:2606.12728},
7 publisher = {{arXiv}},
8 date = {2026-06-10},
9 eprinttype = {arxiv},
10 eprint = {2606.12728 [cs.RO]},
11 shorttitle = {{EquiDexFlow}},
12 url = {http://arxiv.org/abs/2606.12728}
13}