We introduce three architecture each model possesses distinct advantages and achieves state-of-the-art results on the ChicagoFSWild and ChicagoFSWild+ datasets. For more information see our arxiv paper TBA.
Clone and install required python packages:
-
Create env for RGB and KP+RGB
- create conda env with
conda create -n RGB_KP python=3.9 - install torch with
pip install torch==2.0.1 torchvision==0.15.2 --index-url https://download.pytorch.org/whl/cu118 - install turbojpeg with
conda install conda-forge::pyturbojpeg=1.7.7 - install other dependicies with
pip install -r RGB_KP_reqs.txt
- create conda env with
-
Create env for KP
- create conda env with
conda create -n KP python=3.9 - install torch with
pip install torch==2.3.1 torchvision==0.18.1 --index-url https://download.pytorch.org/whl/cu118 - install turbojpeg with
conda install conda-forge::pyturbojpeg=1.7.7 - install other dependicies with
pip install -r KP_reqs.txt
- create conda env with
TBA
| Recoginzers | ChicagoFSWild | ChicagoFSWild+ | Znaki |
|---|---|---|---|
| HandReader_{KP} | 69.3 | 72.4 | 92.65 |
| HandReader_{RGB} | 72.0 | 73.8 | 92.39 |
| HandReader_{RGB_KP} | 72.9 | 75.6 | 94.94 |
We provide models for each architecture trained on datasets ChicagoFSWild, ChicagoFSWild+, and Znaki, which can be downloaded from the link below.
| Pretrained models | Link |
|---|---|
| HandReader_ChicagoFSWild | download |
| HandReader_ChicagoFSWild+ | download |
| HandReader_Znaki | download |
python src/train.py --config-name <cgf_name.yaml>`example to start training model for kp_rgb with dataset Znaki:
python src/train.py --config-name kp_rgb_Znaki.yaml`python src/test_KP_RGB.py --config-name kp_rgb_Znaki.yamlTBA