cosanlab
Customising/Disabling Specific Models
Hi!
Thank you for developing this fantastic toolkit! I am using it to detect facial action units and emotion changes in videos featuring only a single person's face, given that there is only one person in the videos, is it possible to disable the use of the identity_model and would that potentially decrease the processing time?
I am currently processing videos that average around 10 mins in length on an NVIDIA 4090 GPU and extracting per-frame data, any advice on optimising this process would be greatly appreciated.
Hi @irenexu , we are close to completing a major refactor that will allow disabling individual models and also better performance on GPUs. stay tuned.
To speed things up, people have reported better performance by dividing videos into images using ffmpeg and parallelizing these jobs.
