VRDL-HW1

Homework1 in NCTU VRDL

Hardware

The following specs were used to create the original solution.

• Ubuntu 18.04 LTS
• Intel(R) Core(TM) i5-8400 CPU @ 2.80GHz
• NVIDIA RTX 2070

Installation

All requirements should be detailed in requirements.txt. Using Anaconda is strongly recommended. {envs_name} is the new environment name which you should assign.

conda create -n {envs_name} python=3.6
source activate {envs_name}
pip install -r requirements.txt

Dataset Preparation

The training_label.csv is already in the data directory. You can download the data on the Kaggle website: https://www.kaggle.com/c/cs-t0828-2020-hw1/data

Prepare Images

After downloading, the data directory is structured as:

data
  +- training_data
    +- 000001.jpg
    +- 000002.jpg
    ...
  +- validation_data
    +- 000004.jpg
    +- 000005.jpg
    ...

Data Preprocessing

It is going to split the training data randomly to generate a new training data and valid data in the data directory. The ratio of the training data and valid data is 8 : 2

$ python3 preprocessing.py

Training

I provide 2 model for the task. One is ResNet50, and the other is DenseNet201. You can run the ResNet50 model by following

$ python3 ResNet50.py

You can run the DenseNet201 model by following

$ python3 DenseNet201.py

Make Submission

There are two python file to make different submission You can run make_submission_ResNet50 to make a submission for ResNet50 model

$ python3 make_submission_ResNet50.py

You can run make_submission_ResNet50 to make a submission for DenseNet201 model

$ python3 make_submission_DenseNet201.py