NeurIPS 2026 BLV: Synthetic and Real-World Object Recognition for Urban Navigation

A dataset for training and evaluating object recognition and segmentation models on infrastructure relevant to blind and low-vision (BLV) navigation. It contains three configurations, including photorealistic IsaacSim renders for training and real photographs captured by the authors. The dataset was created by NavAble and last updated on May 7, 2026.

Use Cases

• Training object detection models based on photorealistic synthetic renders of urban infrastructure.
• Evaluating model generalization based on the split of real-world photographs for validation and testing.
• Developing segmentation models for urban navigation aids based on infrastructure-focused imagery.
• Pretraining vision models on synthetic data before fine-tuning on real-world photographs.

Strengths

• Includes a canonical test split of real-world photographs for standardized evaluation.
• Provides a mix of synthetic and real-world data configurations for varied training approaches.
• Created by NavAble, an organization focused on navigation accessibility.

Limitations

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

Provenance

Source

NavAble

Collection Method

Likely contains photorealistic IsaacSim renders and real photographs captured by the authors.

Freshness

Last updated 2026-05-07 11:16:56; freshness should be verified.

Geography

Urban environments