RPX: Robot Perception Benchmark with 110,000 RGB-D Frames

RPX is a real-world RGB-D benchmark for evaluating robot perception under embodied deployment conditions, created by IRVLUTD. It includes 100 multi-object scenes and 70 single-object scenes, totaling 110,000 frames. The dataset was last updated on June 3, 2026.

Use Cases

• Benchmarking object detection and segmentation models based on multi-object scene data.
• Training robotic manipulation models using the interaction phase data from multi-object scenes.
• Evaluating 3D scene understanding algorithms based on the provided RGB-D data.
• Developing single-object recognition systems using the 360-degree collections.

Strengths

• Contains 110,000 frames, providing a substantial volume of data for model training and evaluation.
• Includes 100 multi-object scenes, each with three distinct phases (clutter, interaction, clean), simulating real-world conditions.
• Features 70 single-object scenes with 360-degree collections, offering detailed object-level data.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count for the full dataset is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

IRVLUTD

Collection Method

Likely collected through robotic sensors for embodied deployment scenarios.

Freshness

Last updated 2026-06-03 08:01:44; freshness should be verified.