OC-20

The OC20 dataset is a benchmark used to evaluate the performance of SE(3)-equivariant graph neural networks in modeling potential energy surfaces in 3D atomistic systems.

Papers using OC-20 (11)

Equiformer: Equivariant Graph Attention Transformer for 3D Atomistic Graphs2022 · 65 citesReducing SO(3) Convolutions to SO(2) for Efficient Equivariant GNNs2023 · 27 citesThe Open Catalyst 2022 (OC22) Dataset and Challenges for Oxide Electrocatalysts2022 · 14 citesFAENet: Frame Averaging Equivariant GNN for Materials Modeling2023 · 7 citesEwald-based Long-Range Message Passing for Molecular Graphs2023 · 6 citesTransfer learning for atomistic simulations using GNNs and kernel mean embeddings2023 · 5 citesMolecular Geometry-aware Transformer for accurate 3D Atomic System modeling2023 · 3 citesOn the importance of catalyst-adsorbate 3D interactions for relaxed energy predictions2023Uncertainty Quantification in Graph Neural Networks with Shallow Ensembles2025Equivariant Spherical Transformer for Efficient Molecular Modeling2025EquiformerV3: Scaling Efficient, Expressive, and General SE(3)-Equivariant Graph Attention Transformers2026