Mini-NICO Baseline

The baseline is a reference method for the final exam of machine learning course.

Requirements

Installation

we use /python3.7 /torch 1.4.0+cpu /torchvision 0.5.0+cpu for training and evaluation. You can install the pytorch1.4.0 by using this.

conda install pytorch==1.4.0 torchvision==0.5.0 cpuonly -c pytorch

By the way, you can also use the pytorch with cuda to train this baseline.

Prepare Datasets

You need to create the ./data/ folder and put the ./mini_nico/train and ./mini_nico/test in Mini-NICO dataset to the ./data/ directory like

data
├── train
│   └── cat
│   └── cow
│   └──  ..
├── test
│   └── 1.jpg 
│   └── 2.jpg 
│   └──  ..

Split the val data

You can use the following command to split the val data from the train data.

# split the val from the train data and train : val = 7:3
cd utils 
python split_eval_from_train_data.py 

Training

You can use the following command to run for training.

# you can choose the model such as resnet18, resnet34, resnet50, resnet101
python trainer.py --arch=resnet18

If you want to train the method with gpu, you can do this.

# you can choose the model such as resnet18, resnet34, resnet50, resnet101
python trainer.py --arch=resnet18 --gpu

Testing

You can use the following command to run for testing.

# you can choose the model such as resnet18, resnet34, resnet50, resnet101
python test.py --arch=resnet18 --ckpt=your model path

If you want to test the method with gpu, you can do this.

# you can choose the model such as resnet18, resnet34, resnet50, resnet101
python test.py --arch=resnet18 --ckpt=your model path --gpu

After that, you can get the test.csv in the root path ./. And then upload your result to our Mini_NICO_Leaderboard.