Redrecting the implementation of CVPR2022 paper: Transform-Retrieve-Generate: Natural Language-Centric Outside-Knowledge Visual Question Answering web .
Thanks to my fellow @Weizhe Lin, an unofficial implementation of TRiG can be found here: https://github.com/LinWeizheDragon/Retrieval-Augmented-Visual-Question-Answering?tab=readme-ov-file#trig.
I (the first author) certify that this implementation achieves very similar performances of TRiG, following the same method from the official manuscript: pdf
You are also welcomed to implement TRiG by yourself as it is straightforward to understand.
Due to the policies of Amazon, the official implementation of TRiG is not permitted to release (2023/12). Fortunately, you can find the 3rd party implementation from above link for your convenience. Note that, Amazon has not supported or certified the above implementation.