Repository for the Super Resolution project of group 3 for the class of Deep Learning for Visual Recognition, in Aarhus University.
We work with a Conda environment that can be reproduced using the environment.yml
file that we include. To create a Conda environment capable of running the code in this repository, use the following commands.
conda env create -f environment.yml
conda activate PokeRes
If you are running the repo from a server or a computer with a GPU, and you want to train some of the models, you should install environment_cuda.yml
instead.
conda env create -f environment_cuda.yml
conda activate PokeRes
The dataset used for the model is available on Kaggle. Special thanks to kvpratama for sharing it to the open-source community.
The following models are available:
- EDSR: Enhanced Deep Super Resolution, a residual network model proposed on this paper with some minor adjustments
To visualize the super-resolution models that we have obtained, you can use our Streamlit Dashboard app. For building it locally, follow these commands.
streamlit run app.py
There, you can choose the model and the factor of upscaling that you want to visualize, and pick a random Pokemon from the test dataset (which has not been seen by any of the models in the training phases).