SLIC - Semantic Learning for Image Compression

---

The server files are tested only on windows. There might be a few issues running on Linux/Mac because of varying file separator. (Most of the issues can be resolved by updating generate_map.py and combine_images.py)

For reference, Kodak original and compressed images are kept in the code folder.

Requirements

Modules required to test To run the server on local machine, Python 3.6 or above is needed along with the following modules

• flask
• tensorflow
• numpy
• matplotlib
• pillow
• scikit-image
• pandas
• scipy

No other special modules are required for training. Version details for all the modules is available in requirements.txt file

How to run

GUI/Server

• Make sure the trained model files are present in models/ folder
• Make sure the folders and files mentioned in the next section are present.
• Run using the following command

python3 server.py

• The server will be started on localhost:5000

Training

• Make sure you have the dataset downloaded in the data folder along with the pickle files
• Make sure the folders and files mentioned in the next section are present.
• Update params.py, if required
• Run using the following command

python3 train_resnet.py

• The training will start for 200 epochs by default, with learning rate as 0.001

File structure

• models - contains the trained model file
• static - contains some static CSS, JS, image files
• templates - contains the HTML templates for the website
• combine_images.py - methods to encode using JPEG
• frameCapture.py - To test video compression by extracting frames
• generate_map.py - methods to generate heatmap and MS-ROI
• get_metrics.py - methods to calculate PSNR, SSIM
• params.py - params for tarining
• README.md - this file
• requirements.txt
• resnet_model.py - model architecture file
• saveDataNp.py - to improve train performance
• server.py - flask server
• train_resnet.py - to train resnet
• util.py - utils funtions