CSC2516_Project

Github only have the code and the checkpoints cause github doesn't allow to upload files size larger than 100MB.

Dependence

pytorch, keras, opencv, PIL, skimage, sklearn, numpy, tqdm

Train and test with CUDA, haven't tested on CPU.

Below requires the test data

Test data can be downloaded from

https://drive.google.com/file/d/1MpkVXZvtxkKfn0HB33FKk8eiQ-M7rWIa/view?usp=sharing

and put the ./data folder into the root place.

Settings

Default setting is use Regression with skip connection models. Can change to classification models in the following two files by setting args.classification = True or try the models with no skip connection by setting args.skip_connection = False.

train_cifar.py
train_lndscape.py

Each file is for one dataset.

Evaluate with pretrained models

To evaluate the pre-trained models. Default is (regression with skip connection) model.

python train_landscape.py

or

python train_cifar.py

This will return the SSIM score.