Submission: GFCPose/XYZ-IBD/train_pbr

Download submission
Submission name train_pbr
Submission time (UTC) March 9, 2026, 1:05 p.m.
User NUDTCJH
Task Model-based 6D detection of seen objects
Dataset XYZ-IBD
Description only use the train_pbr images for training. 2D detection: YOLO11x
Evaluation scores
AP:0.748
AP_25:0.729
AP_25_mm:0.547
AP_MSPD:0.768
AP_MSSD:0.729
AP_MSSD_mm:0.547
average_time_per_image:-1.000

Method: GFCPose

User NUDTCJH
Publication
Implementation
Training image modalities RGB
Test image modalities RGB
Description

We first predict dense keypoints through a single view, and use these keypoints to match instances across multiple views. We then further refine the keypoints and reconstruct the 3D point cloud through multi-view feature-level fusion of these matched instances. The final pose is solved using the Umeyama algorithm.

Computer specifications 5090