How paul901003/webots_program works
This repository appears to support robotic multi-view capture and visual-hull-based reconstruction of YCB scenes, followed by instance grouping and scene-relation analysis. The principal workflow is to plan viewpoints, capture RGB/depth and pose data, derive masks and labels, carve a hull, and associate voxels into instances. The tree also shows scene-generation, visualization, and evaluation capabilities. There is no README, and sampled source is partial, so links between unsampled supervisors and processing stages are intentionally omitted.
Main entry points: controllers Capture Supervisors · realsense_auto_capture_controller.py Camera Capture · masks.py Mask And Pose I/O · generate_labels.py Rendered COCO Labels
Components 19
Capture And Planning
- Viewpoint Planning
controllers/ycb_supervisor_capture/generate_candidate_viewpoints.py - Capture Supervisors
controllers - UR5e Control
controllers/ur5e_auto_capture_controller/ur5e_auto_capture_controller.py - Camera Capture
controllers/realsense_auto_capture_controller/realsense_auto_capture_controller.py
Masks And Labels
- Rendered COCO Labels
tools/generate_labels.py - Image Segmentation
mobilesamv2/mobilesamv2_seg.py - Mask And Pose I/O
srp/io/masks.py
3D Reconstruction
- Visual Hull Carving
srp/stage1_hull/carve.py - Hull Voxel Data
- Instance Association
instance_hull/associate_voxel.py - Instance Refinement
srp/stage2_instances/merge_bridge_frag.py - Instance Labels
Relations And Evaluation
- Scene Relations
srp/stage3_graph/scene_graph_gt.py - Evaluation And Reports
srp/stage2_instances/eval.py - 3D Visualization
srp/stage2_instances/gen_viz_objs.py
Other components
- Operator
- Webots Simulation
- ROS 2 MoveIt
- Captured RGB Depth Pose
Connections 25
- Operator → Viewpoint Planning: requests views
- Viewpoint Planning → ROS 2 MoveIt: plans paths
- Operator → Capture Supervisors: starts capture
- Capture Supervisors → UR5e Control: commands arm
- UR5e Control → Webots Simulation: moves robot
- Capture Supervisors → Camera Capture: sends capture commands
- Camera Capture → Webots Simulation: reads sensors
- Camera Capture → Captured RGB Depth Pose: saves RGB depth pose
- Operator → Rendered COCO Labels: requests labels
- Rendered COCO Labels → Rendered COCO Labels: renders COCO masks
- Operator → Image Segmentation: requests masks
- Image Segmentation → Mask And Pose I/O: provides masks
- Captured RGB Depth Pose → Visual Hull Carving: supplies views and pose
- Mask And Pose I/O → Visual Hull Carving: supplies masks and pose
- Visual Hull Carving → Hull Voxel Data: writes hull
- Hull Voxel Data → Instance Association: provides occupancy
- Mask And Pose I/O → Instance Association: supplies view masks and pose
- Instance Association → Instance Labels: writes instances
- Instance Labels → Instance Refinement: reads labels
- Hull Voxel Data → Instance Refinement: reads hull surface
- Instance Refinement → Instance Labels: writes refined labels
- Instance Labels → Scene Relations: informs relation analysis
- Instance Labels → Evaluation And Reports: evaluates instances
- Hull Voxel Data → 3D Visualization: visualizes hull
- Instance Labels → 3D Visualization: visualizes instances