Image-Regression---Age-Prediction---Fastai

In the v2 notebook, I have tried to create an Image Regression Model based on Fastai library.

There are few important points to note:

a. Data has been taken from three major sources:

a.1. Wiki IMDB face dataset - https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/

a.2. UTK face dataset - http://aicip.eecs.utk.edu/wiki/UTKFace

a.3. Appa Real face dataset - http://chalearnlap.cvc.uab.es/dataset/26/description/

b. All these three datasets were carefully merged and final model was built on the the merged dataset.

c. Fastai v1 model was used and CNN architecture - ResNet34 was chosen to run the model.

d. In this working notebook, I have used Image Resizing technique in which image sizes were gradually increased which helped in getting higher accuracy.

e. L1 Smooth Loss was used which behaves better than L1 or L2 losses.

f. Learnt using Discriminative Learning Techniques of Fastai in which we can split the NN arch into different parts and assign different values of Weight Decays and Learning Rates for different parts of the NN arch

g. In last, using Fastai Pytorch Hooks and Spotify Annoy was used to create an Image Similarity Model (which did not work very well in my view.

As a testing part of the model, I used India's PM Modi's picture which was taken in 2015 when he was 64 years of age. The model correctly predicted the age at around 63 years.

Hope you enjoy and it may be useful for others.