A manually implemented ResNet-18 network under the PyTorch framework for the CIFAR-10 dataset.
├── data.py # CIFAR-10 dataset loading and preprocessing
├── models.py # Manual implementation of ResNet-18
├── train.py # Training and evaluation pipeline
├── README.md # Project documentation
├── checkpoints/ # Saved model checkpoints
└── logs/ # TensorBoard training logs
The implementation does not rely on any version-specific features and should work with most recent Python and PyTorch releases. The project was developed and tested with:
- Python 3.10.20
- PyTorch 2.12.0 + cu126
- Installation:
conda create -n resnet python==3.10.20
·····
conda activate resnet
pip3 install torch torchvision --index-url https://download.pytorch.org/whl/cu126Download CIFAR-10 dataset and extract it to:
path/to/cifar-10-batches-py
Run the experiment script:
bash run_exp.shTwo experiments will be executed on separate GPUs: one with residual connections and one without.
All training logs are stored in logs/. Launch TensorBoard with:
tensorboard --logdir logs
ResNet18 Training & Validation Loss |
ResNet18 Training & Validation Accuracy |
Plain Network Training & Validation Loss |
Plain Network Training & Validation Accuracy |



