Submission: Drost-CVPR10-3D-Only-Faster/TYO-L

Submission name
Submission time (UTC) Oct. 17, 2019, 7:04 a.m.
User Berti
Task Pose estimation (BOP 2019-2022)
Dataset TYO-L
Training model type Default
Training image type None
Description
Evaluation scores
AR:0.538
AR_MSPD:0.511
AR_MSSD:0.511
AR_VSD:0.594
average_time_per_image:0.747

Method: Drost-CVPR10-3D-Only-Faster

User Berti
Publication Drost et al., Model globally, match locally: Efficient and robust 3D object recognition, CVPR 2010
Implementation
Training image modalities None
Test image modalities D
Description

An implementation of the paper, including ICP for post-processing.

The parameters are set to a "fast" mode (relative sampling distance of 0.05 for model creation and matching).

No images are used, the method uses depth only. The training is done only on the CAD models. The method uses only the CPU, no GPU.

Additionally, the depth range for each dataset is used to limit the search range by thresholding the z-images prior to the search.

Implementation: HALCON 19.05 progress.

Computer specifications Max. 12 Threads on intel Xeon CPU E5-2690 v4 @ 2.60GHz, 2601 Mhz