To do - batch process to csv/sqlite

[robmarkcole]

Table cols are:

• file_path (s3 bucket/key?)
• image_capture_date (if known)
• deepstack_model (face,object, v1 etc)
• label (face if face endpoint else the object)
• confidence
• x_min | y_min | x_max | y_max (us array?)

[robmarkcole]

The schema used on kaggle wheat competition - keep image size and use pixel locations for coordinates. x1 y1 are the center point of the object. Probably nice to do these calculations when we insert into the db. Box size is very useful to identifying anomalies, e.g. giant wheat

[robmarkcole]

Can use datasette or voila for viz ui

• https://datasette.io/plugins