High RAM consumption during custom export

[PMazarovich]

My actions before raising this issue

• Read/searched the docs
• Searched past issues

Steps to Reproduce (for bugs)

1. Create a task with annotations amount, approx. resulting in 380Mb in Cvat for Video
2. Use task_data.group_by_frame() method
3. Monitor cvat_worker_export docker container RAM usage

The thing is that this method - group_by_frame() reads all annotations for task in RAM

Expected Behaviour

Maybe it's better to create a method, that will return Frame for exact frame number, so user could iterate over each separate frame in the task, (so method will fetch data from db and create an individual Frame) without loading all Frames in ram

Current Behaviour

Here is a RAM consumption plot for 75 000 frames. CVAT for video annotation file size is 380Mb

Context

I got this approach from here: https://opencv.github.io/cvat/docs/contributing/new-annotation-format/

Your Environment

CVAT 2.4.5

[bsekachev]

Hi @PMazarovich

Thank you for the report. It is a known issue.

[zhiltsov-max]

It was slightly improved in #6204, which was included in v2.4.7.