|Submission time (UTC)||Aug. 18, 2020, 7:37 a.m.|
|Task||Pose estimation (BOP 2019-2022)|
|Training model type||Default|
|Training image type||Synthetic (only PBR images provided for BOP Challenge 2020 were used)|
|Publication||Pix2Pose: Pixel-Wise Coordinate Regression of Objects for 6D Pose Estimation, ICCV 2019|
|Training image modalities||RGB|
|Test image modalities||RGB|
Poses are estimated using RGB images only without refinement. Results are derived after the following modifications from the original implementation of the paper. Other setups are the same as performed in BOP 2019.
1) Replaced the encoder part with the first three blocks of Resnet-50 with pre-trained weights using ImageNet.
2) Increased a threshold for inlier pixels during PnP-Ransac operation (3->5).
3) A minor bug that causes bad detection results for the T-Less dataset is fixed (different image resolutions were used during training and inference)
4) Increased the number of RPN proposals and NMS thresholds in Mask-RCNN (1000/0.7 to 2000/0.9), which produces more detection proposals
All updates will be shared in our public repository (checkout bop2020 branch after the deadline)
|Computer specifications||CPU: i7-9700K, GPU: Titan V, RAM: 32GB|