This is an unofficial implementation of DOPE (Deep Object Pose Estimation for Semantic Robotic Grasping of Household Objects) trained on self-created synthetic bottle dataset.
$ conda create -n DOPE python=3.6
$ pip install -r requirement.txt
$ conda activate DOPE
[WIP]
- Logs [Download]
DOPE.pytorch
- logs
- Jack_Daniels-checkpoint.pth
- Jose_Cuervo-checkpoint.pth
- Data
DOPE.pytorch
- data
- Real_bottle_sequence
- 000001.jpg
- 000002.jpg
...
- _object_settings.json
- _camera_settings.json
- Run
python demo.py
--path_to_data_dir ./data/Real_bottle_sequence
--class_name Jack_Daniels
--checkpoint ./logs/Jack_Daniels-checkpoint.pth
--plot
- DOPE (Deep Object Pose Estimation for Semantic Robotic Grasping of Household Objects) [paper link]
- Some code is borrowed from below repos.
- Official implementation from nvidia (inference code with ROS) [Deep_Object_Pose]
- Realtime_Multi-Person_Pose_Estimation [pytorch_Realtime_Multi-Person_Pose_Estimation]