Extract landmarks on videos outside the LRW dataset about lipreading_using_temporal_convolutional_networks HOT 4 CLOSED

Hi, you could use dlib library. Our dlib version is 19.17.0

from lipreading_using_temporal_convolutional_networks.

Thanks for the reply. Did you use shape_predictor_68_face_landmarks.dat to extract the landmarks? After extract the landmark did you process it further (i.e., normalize, rotate from side to front)? Thanks.

from lipreading_using_temporal_convolutional_networks.

Hi,

Regrding the shape predictor, we used a model trained on 300W/VW and Helen, but I think you could use the default shape predictor. To my experience, the tracker doesn't make much difference in this case.

Regarding pre-processing, please check the instructions on the readme:
https://github.com/mpc001/Lipreading_using_Temporal_Convolutional_Networks#how-to-prepare-dataset

The first step is to download the tracked points - since you're using your own data you have to track them as mentioned above. Second step is the preprocessing, which points to this script:
https://github.com/mpc001/Lipreading_using_Temporal_Convolutional_Networks/tree/master/preprocessing

Note that you would need to adapt it to run on your own data.

from lipreading_using_temporal_convolutional_networks.

Hi, how should I deal with the situation that dlib do not recognize any face in one frame?

from lipreading_using_temporal_convolutional_networks.