[ICCV'25 Highlight] HccePose (BF): Predicting Front & Back Surfaces to Construct Ultra-Dense 2D-3D Correspondences for Pose Estimation
-
Updated
Sep 5, 2026 - Python
[ICCV'25 Highlight] HccePose (BF): Predicting Front & Back Surfaces to Construct Ultra-Dense 2D-3D Correspondences for Pose Estimation
Publication-ready visualization of 3D objects and point clouds in seconds.
A fork of BlenderProc used in the GRADE framework to generate environments and export some additional information for processing.
DDONGGAE (똥개) — autonomous polyhedron-collecting robot. 1st of 16 teams, AI ROBOT CHALLENGE 2026. ROS 2 Humble, mecanum drive, Nav2-free LiDAR localisation, two-stage YOLO trained entirely on synthetic data.
3D point cloud perception pipeline with synthetic data generation: BlenderProc + Open3D + RANSAC
Synthetic data set builder and object position detector based on Mask-RCNN
One-shot 6D pose estimation for industrial bin-picking, built on SAM-6D. 🏆 Won $3000 for being 5th overall and 2nd among one-shot solutions in the 2025 Perception Challenge for Bin-Picking (OpenCV × Intrinsic).
Synthetic satellite ship imagery generator with YOLO annotations — BlenderProc
To associate your repository with the blenderproc topic, visit your repo's landing page and select "manage topics."