Stick to the work it will yield results. Yesterday i trained a model that was not satisfying as it was overfitting but after investigating i found that the fault is in the data. Most of the faces are in the same allignment so model was predicting the same position everytime still i managed to improve the validation accuracy from 43.23% to 70% and added the work with some sunglasses filters using openCV.
I, Rohit Sethi swear to complete
#100DaysOfCode to the best of my ability and with true spirit.
On Day 6, I did the following :-
Trained a CNN to predict 15 facial KeyPoints and putting filters over faces by detecting faces by haar cascade.
On Day 6, I learnt the following :-
Haar Cascadeto identify bounding box over the face in an image.
Revised - Why
link to the work on Day 6 here