1 of 5

-- Code. Collaborate. Conquer --

DATE: 24 - 25 February 2024

VENUE: Online

Deadline for Registration : 23rd February 2024

Fr. C. Rodrigues Institute of Technology, Vashi, Navi Mumbai

AGNETHON

AGNETHON

2 of 5

Agnethon Team Details

Problem Statement Title: Deepfake Detection System

Team name : Serenity Stars

Domain: AI/ML

Team leader name: Advika Sawant

Member 1: Devishree Nadar

Member 2: Ruvina Vas

Member 3: Sania Joseph

​

3 of 5

Idea /Approach Details

Solution

Objective

-Users should be able to submit suspicious audio, video, and photo files through the platform.

- The system should proactively scan public data and trending content on Instagram and Twitter to assess the authenticity of videos, audios, and photos, particularly in relation to potential deep fake

elements.

- The platform should timely detect deep fake content, contributing to the reduction of misinformation spread across digital platforms, while simultaneously raising awareness about the issue.

- Media professionals and fact-checkers benefit from a robust tool within the platform, empowering them to verify the authenticity of media content in the ongoing fight against

deceptive practices.

​

Users can submit suspicious media files.

​

Deepfake Detection System,Use of AI to detect manipulations in media files.

​

In-depth analysis and verification of submitted files.

​

Continuous Improvement: Regularly updates algorithms for better detection.

4 of 5

Idea /Approach Details

5 of 5

Show Stoppers

  • Voice deepfake detection is not working due to inappropriate meta data, so unable to predict.

​

Dependencies:

  • Deep Learning Frameworks: Utilize TensorFlow, PyTorch, or sklearn for model development.
  • collection and annotating real/fake datasets.
  • Image/Video Processing Libraries,using OpenCV or FFmpeg for data manipulation.
  • Utilizing GPUs or TPUs for efficient model training.
  • Training Model.
  • Utilization standard of metrics for evaluating model performance.
  • Use platforms for project documentation and result reporting.

​

Conclusion

We are able to build deep fake detection system for images and videos.