Adrian Bulat

The memory consumption is likely dominated by the face detection network. You can partially address this by downsampling your images appropriately (note: make sure that you don't downsample to the...

This is likely due to an incompatibility between numba and numpy. Please make sure you are using the latest versions.

The terminologies are largely used interchangeable. This reflects how historically face alignment/key points detection was performed: one would take a deformable facial model and align it with an image such...

Hi @gitjohnnygit The last value is the confidence of the detected bounding box.

While you can add this as a post-processing on top the provided 68 landmarks, this points won't be learned and as such they may or may not correspond with the...

@franz101 Apologies for the delay. It's a matter of having the data to train the models with. If you have a dataset available I can have a look over it....

Hi @Alan-xw the servers are located in UK so depending on your location this may affect your speed. Someone uploaded them here: https://github.com/1adrianb/face-alignment/issues/62 If you still encounter issues, let me...

@lastapple I am really sorry for the inconvenience, depending on your geographical location the download speed will vary indeed. From my tests this appears to mostly be an issue with...

Apologies for the delay, most of the time I go trough the git issue during holidays :) I am afraid that for this models at least here are no other...

Hi Boris, While I haven't run a benchmark (i will run one in the morning and get back to you), at a resolution of 640x480 I get >30fps (I ussually...