search

shaoanlu / face-segmentation-keras

A port of YuvalNirkin/face_segmentation repo to keras.
MIT License
60 stars 28 forks source link

Detecting and cropping the images before inference #3

Open YuvalNirkin opened 6 years ago

YuvalNirkin commented 6 years ago

Hey, Good job converting the original model to Keras!

Please note that the original model was trained following a detection and cropping step. This is different than other segmentation methods which work on full images. Applying the model on full images will yield poor results.

Best, Yuval

shaoanlu commented 6 years ago

Thanks! I'll add your reminder to readme.

jmhummel commented 4 years ago

@YuvalNirkin I really like the results you were able to achieve with the original model, but I'm struggling to determine the specific crop/alignment requirements. I attempted to figure it out by reading your codebase, but I was a bit lost.

Given that I already have the bounding box coordinates and the 68-pt facial landmarks for my images, what are the basic requirements to crop/align the images as required by the model?