We need to deploy a UI to visually validate the results and check what is being returned from the search engine, otherwise, we are either blind (just look to overall P/R/F1) or need to perform manual verifications that comprehend:
- open the DB and check the names of the images and websites returned for a given query
- open each IMG folder and look for the images
- open each TEXT folder and look for the websites.
This is very time consuming. If we have a simple web interface for that, we could also add, e.g., the predictions of the models for each image/website. This way we can also more easily check where the models are failing the most.
We need to deploy a UI to visually validate the results and check what is being returned from the search engine, otherwise, we are either blind (just look to overall P/R/F1) or need to perform manual verifications that comprehend:
This is very time consuming. If we have a simple web interface for that, we could also add, e.g., the predictions of the models for each image/website. This way we can also more easily check where the models are failing the most.