IEEE VIPCup'22 - Top Ten

Oct 8, 2023   | #Deep Learning  #Computer Vision  #Generative AI  #Adversarial Learning  

Competition Title	Vision & Image Processing Cup
Organizer	IEEE International Organization
Year	Dec 2022
Place	Top Ten
Team	Tharindu, Devin, Dilith, Nuwan, Pamuditha, Thieshanthan, Amashi
Code	Paper

Description

In the challenging VIP22 competition, our mission was crystal clear: to distinguish synthetic AI-generated images from natural ones and identify the specific generative model used. This ambitious endeavor led us on a research odyssey through the intricate world of computer vision. From experimenting with generic Vision Transformer-based classifiers to harnessing the power of Contrastive Learning techniques, we left no stone unturned. We even ventured into the realm of adversarial learning models to sharpen our edge in this grand competition.

Technologies

Methodologies

Vision Transformers (ViTs) Contrastive Learning Adversarial Learning

Key Achievements

• Effective Discrimination: We achieved high accuracy in discerning synthetic AI-generated images from their natural counterparts, showcasing our models' robustness in identifying subtle differences.
• Generative Model Identification: Our project excelled in not only detecting synthetic images but also correctly identifying the generative model responsible for their creation.
• Versatile Methodologies: Our wide range of methodologies, from Vision Transformers to adversarial learning, demonstrated the flexibility and adaptability of our approach in complex AI tasks.