Pre-Processing

Resizing the input to 28*28 which is then given as input to model

Data -Info

Classes 3 : Middle , Young , Old
Converted to One-Hot Vector before using

Architecture Used

• Convolution(3,20,5,1) + Relu
• Max_Pool with Stride : 2 and Kernel size :2
• Batch_Normalisation
• Convolution(20,50,5,1) + Relu
• Convolution(50,60,5,1) + Relu
• Max_Pool with Stride : 2 and Kernel size :2
• Unravel (2260 -> 500) + Relu
• Linear (500 -> 10) +Relu
• Linear (10 -> 3)
• CrossEntropyLoss

Optimizer

Stochastic with Momentum is used
Momentum = 0.9
Learning_rate =0.01

Results

Number_epoch : 2 Accuracy : 60.43%
Number_epoch : 10 Accuracy : 73.55%
Number_epoch : 30 Accuracy : 81.21%