Pickapic
Pickapic is an open-sourced application designed for academic research in AI, specifically focused on collecting human feedback on AI-generated images. The tool allows users to generate images from text and then rank them, contributing to the development of a comprehensive text-to-image dataset.
To ensure authenticity and accountability, users are required to authenticate via Google before generating and participating in the project. This authentication process helps maintain the integrity of the data collected.
Pickapic’s mission is to create the largest open-sourced and publicly available human-feedback dataset for text-to-image research. All code, data, and models resulting from the project are made openly accessible to the public. This commitment to transparency encourages collaboration and fosters advancements in AI research.
Currently, Pickapic utilizes the Stable Diffusion 2.1 and ProtogenX3.4 models to generate images from text inputs. These models have been carefully selected to ensure high-quality results and facilitate accurate feedback from users.
The dataset collected through Pickapic will be periodically updated, ensuring that the research project remains up-to-date and relevant. Additionally, the collected data will be released into the public domain, allowing researchers and developers to leverage it for further advancements in AI.
Pickapic is supported by the Google TRC Program, which provides resources and assistance to innovative projects in the field of AI. This support ensures the continued development and improvement of the tool, enabling researchers to make significant strides in the text-to-image domain.