COMPAYL 2024

Inspired by recent advancements in multimodal foundation models, especially vision-language models in pathology, we plan to add a new segment to our workshop: a "Benchside Arena" for vision-language models, echoing the ChatBot arena initiative for large language models.

 

To set up our Beachside Arena, we will collect histopathology images, potentially sourced from the COMPAYL network, and feed them into multimodal models, using both vision and language as input data. Eventually, we aim at making the data used in this experiment publicly available.

 

We will compare their textual outputs, like captions or diagnoses, answers to questions provided as textual prompts, assess results via zero-shot classification tests, challenging the models to identify tissue types from a given set of options, etc.

 

We plan to prepare this experiment in the months before COMPAYL, and present it during the workshop.