Explore the DenseNet deep neural network architecture in this 21-minute video tutorial. Learn about the key differences between DenseNets and ResNets, focusing on the use of concatenation operations instead of identity addition. Discover the benefits of this approach, including improved performance with smaller parameter sizes. Follow along with a comprehensive PyTorch implementation walkthrough, covering dense layers, transition layers, dense blocks, and the complete DenseNet structure. Gain insights into the architecture's ability to alleviate the vanishing-gradient problem, strengthen feature propagation, and encourage feature reuse. Examine the performance improvements achieved by DenseNets on various object recognition benchmark tasks, including CIFAR-10, CIFAR-100, SVHN, and ImageNet.
Overview
Syllabus
- Introduction:
- Background and Context:
- Architecture:
- Data Set:
- Main Results:
- Pytorch Walkthrough:
- High-Level Pytorch API:
- Dense Layer & Transition Layer in Pytorch:
- Dense Block in Pytorch:
- Dense Net in Pytorch:
- Conclusion:
Taught by
Yacine Mahdid
Related Courses
-
Convolutional Neural Networks
-
Stochastic Depth for Neural Networks - Implementation and Analysis
-
Neural Network Programming - Deep Learning with PyTorch
5.0 -
FractalNet Deep Neural Network Explained
-
ImageNet Classification with Deep Convolutional Neural Networks - Paper Explained
-
Convolutional Neural Networks
