Tejas
Gokhale, Assistant Professor in CSEE, delivered a tutorial on "Reliability
of Generative Models in Vision" at the IEEE/CVF Winter Applications of
Computer Vision on Jan 8, 2024.
The tutorial, co-organized with researchers from Arizona State University and University of Maryland College Park, provided a holistic perspective on the challenges and opportunities for developing robust and reliable image generators, given all the discussions around copyright infringements and model compositionality failures.
The tutorial covered the following topics:
1. Recent Advances and Reliability Concerns in Image Generation.
2. Understanding training data memorization in diffusion models and ways to mitigate it.
3. Attribution and Fingerprinting of Image Generative Models.
4. Challenges with Evaluation of Text-to-Image Models.
5. Characterizing and Mitigating the Misalignment Between the Evaluation of Generative Models and their Intended Use Cases
6. Developing Robust Text-to-Image Generative Models in Resource-Efficient Manner.
The tutorial, co-organized with researchers from Arizona State University and University of Maryland College Park, provided a holistic perspective on the challenges and opportunities for developing robust and reliable image generators, given all the discussions around copyright infringements and model compositionality failures.
The tutorial covered the following topics:
1. Recent Advances and Reliability Concerns in Image Generation.
2. Understanding training data memorization in diffusion models and ways to mitigate it.
3. Attribution and Fingerprinting of Image Generative Models.
4. Challenges with Evaluation of Text-to-Image Models.
5. Characterizing and Mitigating the Misalignment Between the Evaluation of Generative Models and their Intended Use Cases
6. Developing Robust Text-to-Image Generative Models in Resource-Efficient Manner.
The tutorial was well-received by the
WACV audience, who appreciated the breadth and depth of the topics covered, as
well as the practical insights and recommendations provided by the
speakers.
More information is available at https://asu-apg.github.io/rgmv/. For more information, please contact
Tejas Gokhale at gokhale@umbc.edu.