/Implementing-DenseNet-research-paper-on-cifar-10-dataset

Primary LanguageJupyter Notebook

This repository conatins a notebook which implements Research paper : Densely Connected Convolutional Networks proposed by Gao Huang at CVPR 2017 under subject Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG).

Link to the researche paper : https://arxiv.org/abs/1608.06993

Screenshot (253)

DenseNets have several compelling advantages:

  1. They alleviate the vanishing-gradient problem.
  2. Strengthen feature propagation.
  3. Encourage feature reuse.
  4. Substantially reduce the number of parameters.

Results: We did not used any dropout layers or any other regularization techniques at all and without overfitting got a validation accuracy of 0.9 on benchmark CIFAR10 dataset.