Application supporting passport photos creation with use of Deep Neural Networks. Application:
- detects face -> ML Kit Face Detection,
- verifies if the head is oriented straight, eyes are open and facial expression is neutral -> ML Kit Face Detection,
- segments person from the background -> TensorFlow lite munet model,
- verifies if background is plain and bright -> OpenCV,
- verifies if face does not contain shadows -> TensorFlow, OpenCv,
- TBD: verifies if eye pupils are directed towards camera,
- verifies if there are no objects partially covering face -> OpenCV,
- after photo capturing attempts to improve face shadows -> TensorFlow lite GAN pix2pix model. See below!
- after photo capturing attempts to improve background -> OpenCv,
- cuts the photo to correct format and saves it to the phone,
- if chosen, taken picture is saved 8 times on 15x10 cm paper for convinient printing.
For enabling shadow removal please download a shadow removal model from here and place it in assets folder. Usage of the application without this model is possible, but shadows will not be corrected. Shadow removal model was taught using pix2pix. Training data was generated with help of DPR and face-seg. Code for data generation can be found HERE.
Consecutive previous of the application
