This is my personal project in Introduction of Artificial Intelligence.

This project is about image classification about ResNet(slightly different) and FC on Cifar-10 dataset.

-This code was built on Ubuntu 16.04 with Anaconda Python 3.6 and PyTorch v1.1.

-NVIDIA GPU is needed for both training and testing.

-I used RTX 2080 for training and testing.

0. [Recommended] Create a new conda environment.

   It's recommended if you change your anaconda source to Tsinghua source.

   conda create -n cifar10 python=3.6
   

   And activate the environment.

   conda activate cifar10
   

1. Install pytorch.1.1.

   conda install pytorch=1.1 torchvision cudatoolkit=10.0
   

   Or from source

   conda install pytorch=1.1 torchvision cudatoolkit=10.0 -c pytorch
   

2. Install dependencies.

   pip install matplotlib yacs tqdm
   

   Or from Tsinghua source:

   pip install -i https://pypi.tuna.tsinghua.edu.cn/simple matplotlib yacs tqdm
   

3. Create directories for data and models:

   mkdir data
   mkdir work_dirs

Train model named fc3 whose network is fc-3 with batchsize=400

python train.py batch_size 400 model fc3 network fc_3

Train model named res101 whose network is res-101 with batchsize=200

python train.py batch_size 200 model res101 network res_101

Test model res101

python train.py --test batch_size 400 model res101 network res_101

Visualize outputs of model res101

python train.py --test --vis batch_size 400 model res101 network res_101

The output of visualization will be saved in work_dirs/model_name/vis/