Method: DPOD (synthetic)

User ivan_shugurov
Publication Sergey Zakharov, et al. DPOD: Dense 6D Pose Object Detector in RGB images, ICCV, 2019
Training image modalities RGB
Test image modalities RGB

One network per scene was trained. No refininement was used.

For the scenes which include multiple instances of the same object, object boundaries were additionally predicted.

For the linemod objects the UV maps described in the paper were used, for the other datasets normalized 3D coordinates were used instead.

The networks were trained on synthetic data. For training on the TLESS dataset, only the provided textureless CAD models were used.

Computer specifications Max. 16 Threads on Intel Core i9-9900K, RTX 2080 Ti

Public submissions

Date Submission name Dataset
2019-10-10 13:03 - LM-O
2019-10-10 13:36 - HB
2019-10-10 13:45 - TUD-L
2019-10-10 14:22 - YCB-V
2019-10-12 13:44 with instance segmentation T-LESS
2019-10-12 14:11 - IC-BIN
2019-10-13 11:22 - ITODD
2019-10-14 09:10 - LM