Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

Learning Dense Hand Contact Estimation from Imbalanced Data

Authors: Daniel Jung, Kyoung Mu Lee

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental 5 Experiments 5.1 Datasets 5.2 Evaluation metrics 5.3 Ablation study 5.4 Comparison with state-of-the-art methods
Researcher Affiliation Academia 1IPAI, 2Dept. of ECE & ASRI, Seoul National University, Korea EMAIL
Pseudocode No The paper describes the methodology in prose and figures, but does not include any explicitly labeled pseudocode or algorithm blocks.
Open Source Code Yes The codes are available at https://github.com/dqj5182/HACO_RELEASE.
Open Datasets Yes We select 14 datasets with diverse hand interactions including Ob Man [19], Dex YCB [5], MOW [4], HO3D [15], H2O3D [16], ARCTIC [12], HOI4D [36], H2O [24] for hand-object interaction, Inter Hand2.6M [41], HIC [59] for hand-hand interaction, PROX [17], RICH [22] for hand-scene interaction, and Decaf [55], Hi4D [65] for hand-body interaction.
Dataset Splits Yes To reduce redundancy from large video datasets, we employ sampling ratio of 5, 10, 5 for HOI4D [36], Inter Hand2.6M [41], and Decaf [55] dataset, respectively. ... For the RICH dataset, we follow the official split used in BSTRO [22] for fair comparison.
Hardware Specification Yes We train HACO for 10 epochs on a single NVIDIA A6000 GPU.
Software Dependencies No Py Torch [47] is used for implementation. Our backbone is initialized with the pre-trained weights of publicly released Ha Me R [49].
Experiment Setup Yes We use the Adam W optimizer [38] with a learning rate of 10 5 and a mini-batch size of 24. For stable convergence, the learning rate is reduced by a factor of 0.9 after 5 and 10 epochs. We train HACO for 10 epochs on a single NVIDIA A6000 GPU.