signapse 1.0.19

Signapse_synthetic_signer

Signapse synthetic signer

Tested using python=3.10.9

Example command

python inference.py "{\"GPU\":\"cpu\",\"input_video_path\":\"path/to/video.mp4\",\"input_type\":\"video\",\"start\":0,\"stop\":-1,\"num_workers\":1,\"GAN_folder\":\"GAN\",\"signer\":\"Barbara\",\"stereo\":\"False\",\"pro\":\"True\",\"detailed_video\":\"False\",\"organisation_id\":\"0\",\"sign_request_id\":\"0\",\"sign_result_queue_url\":\"0\",\"synthetic_sign_results_bucket\":\"0\"}"

GAN_Checkpoint files

GAN input files need to be downloaded to the respective folders. The GAN files are saved on.

To choose the output signer appearance, pass the respective --signer XX argument.

Barbara

--signer Barbara

Download the folder and store as ./Barbara.

Safyan

--signer Safyan

Download the folder and store as ./Safyan.

Mia

--signer Mia

Download the folder and store as ./Mia.

Face and Body Separate GANs

To run without Face and Body (--GAN_folder GAN)

Pass the respective --GAN_folder GAN argument.

--GAN_folder GAN

To run with Face and Body (--GAN_folder GAN_FaB)

Pass the respective --GAN_folder GAN_FaB argument.

--GAN_folder GAN_FaB

Face Mesh

To run with FaceMesh and FaceLandmarks (--GAN_folder GAN_FM)

--GAN_folder GAN_FM

To run with FaceMesh and no FaceLandmarks (--GAN_folder GAN_FM_NoFL)

--GAN_folder GAN_FM_NoFL