Text to Image Generative Model
Text-to-image generative models synthesize realistic images from textual descriptions, aiming to bridge the gap between human language and visual representation. Current research heavily focuses on addressing biases inherent in these models (often stemming from training data), improving compositional accuracy and mitigating vulnerabilities to adversarial attacks, including poisoning and backdoor attacks. These models are rapidly impacting various fields, from art and design to scientific visualization, but their ethical implications and potential for misuse necessitate ongoing investigation into robustness, fairness, and accountability.
Papers
Nightshade: Prompt-Specific Poisoning Attacks on Text-to-Image Generative Models
Shawn Shan, Wenxin Ding, Josephine Passananti, Stanley Wu, Haitao Zheng, Ben Y. Zhao
Localizing and Editing Knowledge in Text-to-Image Generative Models
Samyadeep Basu, Nanxuan Zhao, Vlad Morariu, Soheil Feizi, Varun Manjunatha