Workflow
AI‑assisted asset generation with human QC, followed by scalable scene creation and automatic labeling.
-
1) Asset Generation
- Concept → AI placeholders → human redesign & scaling
- PBR texturing via Quixel/Substance; QC scripts for topology and pivots
- Metadata tags and multi‑format export
-
2) Scene Creation
- Base assembly with library assets; procedural variation (≈10/scene)
- Lighting & baking; plausibility and coverage review
-
3) Annotation & Delivery
- RGB, depth, normals, segmentation, instance masks, boxes
- COCO JSON, YOLO TXT, KITTI, Pascal VOC; USD/FBX/GLTF/BLEND
Output Structure
/scene_name/ /images/ scene_variation01.png scene_variation02.png /annotations/ /bounding_boxes/ coco_annotations.json yolo_labels/ /segmentation/ semantic_masks/ instance_masks/ /depth/ /normals/