gdsc-iiit-kalyani / fusion_vae_dsc Goto Github PK

The core of the project relies on using Variational Auto-encoders to "fuse" features from two images into a new one. To those familiar with Genetic Algorithms, similar thing is tried here. Currently in code is a simple "fusion" wherein we arbitrarily merge features from two images. A better "fusion" algorithm/suggestion is desirable.

Note: This issue may be assigned to multiple assignees and will remain open. PRs referencing this issue and improving on the previous best algorithm would be merged, allowed other assignees to improve upon the same. As a starting point, we recommend reading up on VAEs and GAs and see possible combinations. If implementing an idea borrowed in portion or in its entirety from somewhere, we expect the PR to have permissions of using those ideas/algorithms as well as to cite all relevant sources.

This is a request to the maintainers to kindly put up a few sample issues for beginners so they get an idea of how someone can contribute to this repository, probably hinting on what possible implementations could resolve the issue as well.

The Jupyter notebook is v1.0 of the code initially planned. It must be refactored and properly documented. It is desirable to port the code into .py files and properly document it. Assignees should not make changes to the logical structure for the code in the PR referencing this issue. For any logical changes, please open a new issue and reference a PR.

The code currently lacks testing and an automated pipeline. Assignees are expected to perform all or some of the tasks identified below:

• Make code modular

• Add exception handling wherever applicable to prevent unintentional clashes

• add unit tests and sanitize inputs to functions if they appear to cause crashes. If sanitization is not possible, report to the developer

• put the model on Tensorflow serving

• add CI pipeline

The images displayed in README.md has an excessive amount of unused white space. Crop the main part of each image and add the labels 'Parent - 1' ,'Parent - 2' and 'Child' to each image. Make sure not to make any changes to the image itself ,Other than cropping and adding text.