OpenLex3D: A New Evaluation Benchmark for Open-Vocabulary 3D Scene Representations

Christina Kassab, Sacha Morin, Martin Büchner, Matias Mattamala, Kumaraditya Gupta, Abhinav Valada, Liam Paull, Maurice Fallon

Keywords: 3D Scene Representations, Open Vocabulary Evaluation, Instance Retrieval, Instance Segmentation

Abstract: We present a new evaluation benchmark for open-vocabulary scene representations, generating novel label sets for widely used RGB-D datasets. This enables detailed analysis of failure cases and potential improvements. The benchmark is publicly available at: https://openlex3d.github.io.

Submission Number: 2