• Upload Models to Dropbox and update the download model link
• Upload Prototxt Files
• Write unified evaluation script.

This repo contains the caffe models, prototoxt files and the evaluation scripts from Camgoz et al's "SubUNets: End-to-end Hand Shape and Continuous Sign Language Recognition" (ICCV'17).

• Download and compile the custom caffe-ctc developed by ChWick including pyCaffe.
• Download and extract One-Million-Hands dataset.
• Download and extract RWTH-PHOENIX-Weather 2014: Continuous Sign Language Recognition Dataset dataset.
• Download and install TensorFlow 1.3.0+ (TF is used for beam search decoding. Tested with TF 1.3)
• Download models using the script Models\download_models.sh

python evaluate.py --model <model_name> --phoenix_dir </path/to/phoenix/2014/> --one_mil_dir </path/to/one/mil/hands/>

Models which require phoenix_dir to be set

• WordSubUNet-FullFrame-SingleLayer
• WordSubUnet-FullFrame-TwoLayer
• WordSubUnet-HandFrame-TwoLayer
• SignSubUNets-Expert
• SignSubUNets-Generic

Models which require one_mil_dir to be set

• HandSubUNets

Please cite the paper below if you use this code in your research:

@inproceedings{camgoz2017subunets,
  author = {Necati Cihan Camgoz and Simon Hadfield and Oscar Koller and Richard Bowden},
  title = {SubUNets: End-to-end Hand Shape and Continuous Sign Language Recognition},
  booktitle = {IEEE International Conference on Computer Vision (ICCV)},
  year = {2017}
}