In this video I go through "Aggregated Residual Transformations for Deep Neural Networks" paper and implement it in PyTorch. The key idea of the paper is to keep the complexity and improve accuracy of the model. They did it by combining ResNet/VGG and Inception into new model called ResNeXt.
Paper:
[ Ссылка ]
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
GitHub Repo:
[ Ссылка ]
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Connect with me on:
Linkedin - [ Ссылка ]
GitHub - [ Ссылка ]
Medium - [ Ссылка ]
Buy Me a Coffee - [ Ссылка ]
▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬▬
Timestamps:
0:00 Paper Overview
0:33 Inception and ResNet/VGG
2:38 ResNeXt Block
3:20 Group Convolution
5:05 Architecture
5:50 Results/Implementation Details
8:18 ConvBlock
9:00 ResNeXt Block
14:35 Overall Architecture
15:33 Testing & Fixing
![](https://i.ytimg.com/vi/CANodHhCyCw/maxresdefault.jpg)