Closed steventkrawczyk closed 10 months ago
Super cool. @steventkrawczyk , @NivekT mind assigning me? Current thought process is to build the experiment to measure against basic metrics like Peak signal-to-noise ratio
, Structural Similarity Index
, Signal to reconstruction error ratio
. I think we may need to carve out a separate experiment class for image comparison (possibly a third for video comparison as well?) since the current experiment class is built to handle text with heavy support for dataframes.
Sure, feel free to start on that. If anyone else has a different idea, please feel free to comment here as well.
Thanks @HashemAlsaket !
🚀 The feature
We should look into experiments and eval functions for image models, like stable diffusion
Motivation, pitch
Many GenAI apps are multi modal, and we'll have to support more than LLMs in the long term
Alternatives
No response
Additional context
No response