Keypoint RCNN based robot joints detection. Original implementation is from Mask R-CNN:
[Test image source]: Deep Robot-to-Camera Extrinsics for Articulated Manipulators (DREAM)
Note
This repository has been tested on Google Colab. It also depends on opencv-python, albumentations, torch, and torchvision:
git clone https://github.com/kidpaul94/Robot_Keypoints.git
cd Robot_Keypoints/
pip3 install -r requirements.txt
Note
custom_train.py
receives several different arguments. Run the --help command to see everything it receives.
python3 custom_train.py --help
Note
eval.py
receives several different arguments. Run the --help command to see everything it receives.
python3 eval.py --help
{"bboxes": [[x_c, y_c, w, h]], "keypoints": [[[x1, y1, v1], [x2, y2, v2], [x3, y3, v3], ...]]}