A Pytorch implementation of a network based on Residual Blocks with computational cost in mind. It outputs the raw logits so it can be used for classification, regression or feature embedding tasks.
The development was made in Python
3.10
and torch
2.xx
.
The main network is the smaller implementation with 15 layers, but it is prepared to assume other sizes depending on the network's building parametrization.
- Implement mResNet and test for binary classification.
- Add Dockerfile
- Finish README.md
- Share training and testing notebooks
- Add requirements.txt
- Run tests comparing to MobileNet
For citation purposes, please refer to this repository.