Click here to view the synthetic dataset generation demo video
The core objective of this repository is to delve deep into the realm of Generative AI, focusing primarily on its capability to generate synthetic datasets. This is especially beneficial in scenarios where traditional data collection is not only challenging but potentially impossible for instance- defect data, and confidential data. Through advanced Generative AI techniques, our mission is to bridge this gap and provide solutions where creativity and synthesis in data generation are applicable.
In essence, this repository may serve as a repository for researchers and innovators to modify, fine-tune, and explore the proposed/ applied approaches further. Our aim is to produce more diverse synthetic datasets that can replicate the complexity of real-world data landscapes.
- Research Papers: A curated collection of papers and articles pivotal to understanding the intricate layers of Generative AI, especially in the context of complex data generation.
- Image Data Augmentation Approaches: A Comprehensive Survey by Kumar, Teerath et al., 2023
- Text Data Augmentation for Deep Learning, Journal of Big Data, 2021
- A survey of automated data augmentation algorithms for deep learning, Knowledge and Information Systems, 2023
- Defect Transfer GAN: Diverse Defect Synthesis for Data Augmentation by Wang, Ruyu et al., 2023
- Defect-GAN: High-Fidelity Defect Synthesis for Automated Defect Inspection by Zhang, Gongjie et al., 2021
- An Introduction to Variational Autoencoders by Kingma, Diederik P. and Welling, Max, 2019
- Diffusion Priors In Variational Autoencoders by Wehenkel, Antoine and Louppe, Gilles, 2021
- Improving VAE based molecular representations for compound property prediction, Journal of Cheminformatics, 2022
- Synthetic Data Generation for Steel Defect Detection and Classification Using Deep Learning, Symmetry, 2021
- Segmentation-Based Deep-Learning Approach for Surface-Defect Detection, arXiv preprint arXiv:1903.08536, 2019
- Defect classification on limited labeled samples with multiscale, Applied Intelligence, 2021
- A survey of defect detection applications based on generative adversarial networks, IEEE Access, 2022
- Synthetic data augmentation for surface defect detection and classification using deep learning, Journal of Intelligent Manufacturing, 2022
- Synthetic Defect Generation for Display Front-of-Screen Quality Inspection: A Survey, arXiv preprint arXiv:2203.03429, 2022
- Deep-learning-based computer vision system for surface-defect detection, Computer Vision Systems: 12th International Conference, ICVS 2019, Thessaloniki, Greece, September 23--25, 2019, Proceedings 12
-
Code: Colab Notebook, Reproducible code segments, tailored for various Generative AI models optimized for creating synthetic datasets.
-
Datasets: References to existing datasets and examples of synthetic datasets generated using the methods proposed herein. COMING SOON
-
Tutorials: Comprehensive guides and walkthroughs demystifying the technical facets of Generative AI for dataset creation. COMING SOON
--Published Papers Scholarly contribution by our team . The latest paper that used GenAI based approach to handle the limited sample issues and applied transfer learned ViT models for the classification - https://arxiv.org/abs/2401.00393
5: The paper where we found potential and drawbacks of transfer learned ensembled CNN and recommended transfer learned ViT - https://arxiv.org/abs/2311.14824
4: The survey where we found the gaps, challenges, and requirements to resolve the scarce data - https://ieeexplore.ieee.org/abstract/document/10292659
2-3: The preliminary contributions where we came up with the conceptualized framework were limited only to overfitted binary classification-
The RailTwin Framework was introduced here: https://ieeexplore.ieee.org/abstract/document/9926529
The necessity of reusable models was realized here: https://ieeexplore.ieee.org/abstract/document/9926693
1: An extra book chapter: It talks about AI in DT for health and well-being; but we could use the takeaways in our next research because identifying defects was also a diagnostic problem. https://www.sciencedirect.com/science/article/abs/pii/B9780323991636000111
- Clone the Repository:
git clone https://github.com/turna1/GenAI.git
- Copy and edit the Colab notebook directly:
Your insights are valuable! If you're enthusiastic about expanding the current state of Generative AI for dataset creation, we're all ears. This could be in the form of new research papers, enhanced code models, dataset examples, or tutorial improvements. To begin contributing, please go through our Contribution Guidelines.
This project operates under the MIT License. For more details, please see LICENSE.
For collaborations, queries, or suggestions, please contact:
- Name: Rahatara Ferodusi
