# Digital Image Processing Final Project(R12631042)
___

## Research Topic :
#### Use The IMDB-WIKI dataset 500k+ face images with age and gender labels for deep learning training to accurately predict the age of a single image

___
## Research Object：

#### *Evaluation protocol： $\epsilon= 1-\exponential^\frac{-(𝒙-𝜇)^2} {(2𝜎^2)}$*

### I hope that my testing sets performance can achieve an error of less than 0.3
___

## Research Methods：
* #### Use the VGG-16 architecture for convolutional neutral networks
* #### Pretrained on ImageNet for image classification
* #### The 524,230 face images crawled from IMDB and Wikipedia websites
* #### Pose the age regression as a deep classification problem followed by a softmax expected value
* #### The CNNs finetune on the IMDB-WIKI dataset, and then on LAP images
* #### When training for classification, the output layer is adapted to 101 output neurons corresponding to natural numbers from 0 to 100, the year discretization used for age class labels

## Detailed Implementation Steps:
#### 1. observe the data type

## References :
* #### https://data.vision.ee.ethz.ch/cvl/rrothe/imdb-wiki/
* #### DEX: Deep EXpectation of apparent age from a single