This is the implementation of Real-time Pose Estimation from Images for Multiple Humanoid Robots, RoboCup Symposium 2021.
The HumanoidRobotPose dataset provided in this repository also can be downloaded as zip file from here. The dataset format is the same as the COCO dataset.
To use the dataset, install pycocotools:
pip install -U 'git+https://github.com/cocodataset/cocoapi.git#subdirectory=PythonAPI'
To train the model, run:
python src/main.py --config experiments/train.yaml --output_dir output_path --dataset_path dataset/hrp/
To evaluate the model, run:
python src/main.py --eval --config experiments/test.yaml --dataset_path dataset/hrp/ --checkpoint checkpoint_path
The pretrained model is available here.