CNN + Transformer

TODO

1. ResNet 50, 101 train / test able code (MNIST)
2. ResNet 50, 101 train / test able code (COCO)
3. Transformer train / test able code (COCO)
4. ResNet + Transformer (COCO)

Usage

An example of training/testing ResNet50 or ResNet101

CUDA 11.0 with CUDNN 8.2.0

0) Prepare Python 3.7 virual environments and set requirements.txt

conda create -n [env name] python=3.7 -y
git clone https://github.com/wlgjs8/Transformer
cd Transformer
pip install -r requirements.txt

1) Train Model on MNIST Dataset

python example/train_resnet.py -net resnet50

2) Test Model on MNIST Dataset

python example/test_resnet.py -net resnet50 -weights [path to checkpoint]

Directory Structure

├── coco
│   ├── train2017
│   ├── test2017
│   ├── val2017
│   └── annotations
├── data
│   ├── MNIST
│   └── cifar-100-python
├── checkpoint
    └── resnet

Dataset

https://drive.google.com/file/d/1--foZ3dV5OCsqXQXT84UeKtrAqc5CkAE/view