raman-raje/densenet-of-cifar-10

DenseNet-of-CIFAR-10

Densenet-Tensorflow

Keras implementation of Densenet using Cifar10

• The code that implements this paper is present in IPYNB

If you want to see the original author's code or other implementations, please refer to this link

Requirements

• Tensorflow 1.x

What is the "Dense Connectivity" ?

What is the "Dense Block" ?

What is the "Densenet Architecture" ?

Compare Structure (CNN, ResNet, DenseNet)

Implementation:

• Implemented DenseNet-BC (k =12 ,L = 100) [0.76M parameters].
• Network Trained on 70 Epochs.
• No Dropout was used
• Image Augmentation was used.

Results:

I was able to achieve accuracy of 92.7% by training for 70 Epochs.

Loss And Accuracy Plots

Confusion Matrix