LipSick Pre-trained Models

This is the default DINet model "clip_training_DINet_256mouth.pth" renamed "pretrained_lipsick.pth" to work with adjusted paths. The model should be place in ./asserts

This release also contains the Deepspeech model named output_graph.pb and should also be placed in ./asserts along with pretrained_lipsick.pth

This release also contains shape_predictor_68_face_landmarks.dat which is used by Dlib for face tracking so we have landmarks, the model should be placed in ./models

Note: The released pretrained model "pretrained_lipsick.pth" is trained on HDTF dataset with 363 training videos so the generalization is limited,
credit of creation goes to the original trainers Zhimeng Zhang & associates.