Can I download training data for my own Machine Learning classifiers?

Yes. Download all the high resolution motion-region croppings automatically by using a date range query with the search operator:

filetype:cropping

For example, to download all cropped images that were labeled "car" from 2018-08-16T13:56:48.553-0700 to now, you enter this query:

cars 2018-08-16T13:56:48.553-0700 to now filetype:cropping

Press the Download script button in the Search Box panel:

download_croppings_Screen_Shot_2018-08-16_at_5.04.24_PM.png

You'll receive an email with a link to download the shell script that fetches all the croppings from events that match your query. When you run the script, the output will look something like this as each file is downloaded via curl:

$ bash camio-download.sh /tmp/camio-20180816 ICR-310WS-2018-08-16T20:56:58.633-0000.jpg % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 390k 100 390k 0 0 502k 0 --:--:-- --:--:-- --:--:-- 519k ICR-310WS-2018-08-16T20:56:54.712-0000.jpg % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 105k 100 105k 0 0 287k 0 --:--:-- --:--:-- --:--:-- 288k ICR-310WS-2018-08-16T20:56:54.712-0000.jpg ICR-310WS-2018-08-16T20:56:51.829-0000.jpg % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 8203 100 8203 0 0 8543 0 --:--:-- --:--:-- --:--:-- 8571 ICR-310WS-2018-08-16T20:58:11.945-0000.jpg % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 102k 100 102k 0 0 306k 0 --:--:-- --:--:-- --:--:-- 308k ICR-310WS-2018-08-16T20:58:05.346-0000.jpg % Total % Received % Xferd Average Speed Time Time Time Current Dload Upload Total Spent Left Speed 100 54277 100 54277 0 0 161k 0 --:--:-- --:--:-- --:--:-- 163k ICR-310WS-2018-08-16T20:58:05.346-0000.jpg

The resulting directory (e.g. "camio-20180816" in this case) contains all the images organized by camera and timestamp:

After you train your classifiers, you can register them as Hooks to perform analyses and add labels to the Event. If your classifier applies only to certain types of Events, then you can reduce the callbacks to your servers by specifying a query_match that filters the Events that trigger your analysis Hook (e.g. the query_match "cars" ensures that only Events with cars trigger your Hook).

 

NOTE: currently the filetype:cropping search operator fetches only high resolution croppings computed for Box-connected video streams using advanced object labeling, text recognition, or face recognition subscription plans. Camio will soon fetch them regardless of subscription plan.

Have more questions? Submit a request

Comments