《STPose：Sparse attention and cross-layer connected transformer networks for 6D pose estimation in the power industry》

System

Our automated labeling system code is available:
1.The system is a pure python environment without compiling.
2.The system can be used to annotate the 6D pose estimation data set of objects composed of video sequences. Annotators only need to manually annotate the 6D poses of objects in the first frame of each video sequence, and the annotations of subsequent frames can be automatically calculated and generated by the system.
3.Run the label_tool.py file to start the system, and click the auto label button to start automatic labeling.
4.The BA_Optimization folder stores codes related to BA optimization. At present, the effect of BA optimization is poor, and the follow-up needs to be improved.

Dataset 

The dataset , which is a dataset for power devices, built using this system is available at:
https://pan.baidu.com/s/1iu-55rP5BZQ4JaAApHnKZQ
pwd=jm3r
The format of the dataset is as follows：
├── train.json
├── test.json
├── camera.json
├── test_targets_bop19.json
├── image_lists/
├── detector_annotations/
├── test_targets_bop19.json
├── train/000000、... 、000080/
│ ├── rgb/
│ ├── mask_visib/
│ ├── mask/
│ ├── depth/
│ ├── scene_gt_info.json
│ ├── scene_gt.json
│ ├── transform.npy
│ ├── scene_camera.json
├── test/000000、... 、000020/
│ ├── rgb/
│ ├── mask_visib/
│ ├── mask/
│ ├── depth/
│ ├── scene_gt_info.json
│ ├── scene_gt.json
│ ├── transform.npy
│ ├── scene_camera.json
├── model
│ ├── models_info.json
│ ├── obj_000001.ply~obj_000016.ply

Method

Our STPose network code is being organized and will be published soon.