todo.md

TODO:

• Fix sample images
• Allow other image types
• Allow the model to iteratively sample text
• Add nucleus size and other advanced options

Please note that this model was explicitly not trained on images of people, and as a result is not designed to caption images with humans. This demo accompanies our paper RedCaps Created by [Karan Desai](mailto:kdexd@umich.edu), [Gaurav Kaul](mailto:kaulg@umich.edu), [Zubin Aysola](mailto:aysola@umich.edu), [Justin Johnson](justincj@umich.edu)