Multi-view 3D Referring Expression Segmentation Benchmark

MVRefer is the first benchmark dataset for Multi-view 3D Referring Expression Segmentation (MV-3DRES). It was created by sosppxo and last updated in February 2026 to align 3D language grounding with the realistic sensing conditions of embodied agents.

Use Cases

• Train models for 3D language grounding using multi-view RGB images and referring expressions.
• Benchmark segmentation performance for embodied agents perceiving environments through casually captured views.
• Develop algorithms for Multi-view 3D Referring Expression Segmentation (MV-3DRES) tasks.

Strengths

• First benchmark dataset specifically designed for the MV-3DRES task.
• Built to align with realistic sensing conditions of embodied agents using casually captured RGB views.
• Last updated in February 2026.

Limitations

• Dataset size, row count, and column details are unknown.
• Specific file formats and sample data are unavailable for review.
• Geographic and temporal coverage of the collected scenes are not specified.

Provenance

Source

huggingface

Collection Method

Built for the MV-3DRES task, aligning 3D language grounding with realistic sensing conditions from embodied agents.

Time Range

null

Freshness

Last updated 2026-02 -27.

Geography

null