STRIPES: Synthetic Warehouse Object Detection Dataset

Synthetic dataset of 30,000 object detection images across three warehouse layout categories (ground, shelf, trailer) with COCO format annotations.

Total: 1,001,228 annotations
Object class: box (single-class detection)

Generated via Blender 3D simulation for "Curating the Damaged and Cluttered Reality: Generative and Physics-Enabled Simulation for Sim-to-Real Warehouse Perception", published in the proceedings of the Curated Data for Efficient Learning (CDEL) Workshop at the European Conference on Computer Vision (ECCV) 2026.

All 1,001,228 annotations include both bounding box AND instance segmentation data (100% complete). Each object annotation contains:
- Axis-aligned bounding box [x, y, width, height]
- Precise instance segmentation mask (polygon coordinates)
- Pixel area and category information

Files:
- ground/images/ (10,000 JPEG files)
- ground/instances.json (464,787 annotations)
- shelf/images/ (10,000 JPEG files)
- shelf/instances.json (275,821 annotations)
- trailer/images/ (10,000 JPEG files)
- trailer/instances.json (260,620 annotations)

No proprietary or confidential Zebra Technologies data included. Dataset contains only synthetic renders suitable for domain adaptation research.