Incorrect data returned in frames meta request for a ground truth job

[bsekachev]

Actions before raising this issue

• I searched the existing issues and did not find anything similar.
• I read/searched the docs

Steps to Reproduce

1. Create a task with Ground Truth job, using attached archive (Frame selection method: random).

Content of the image corresponds to the image number in this archive). E.g. this is image_10.jpg. Each image resolution is random. This specific image has resolution 800x900.

2. Open GT job and check meta response. It has resolution 1500x900. The image distorsed

Of course, image name also may be incorrect:

images.zip

Expected Behavior

No response

Possible Solution

No response

Context

No response

Environment

commit db293b7024ff9252cfb8bb5c648364539d4a6f09

[O-J1]

I am observing this issue as well as a paying customer on the cloud service and all of my ground truth images display a name of 'placeholder.jpeg' and most are distorted.

[zhiltsov-max]

The situation is the following:

1. /api/tasks/<id>/meta returns all the task frames for simple GT jobs (it has always been like this), the excluded frames have dummy information in the "frames" field
2. For GT jobs in consensus tasks the logic is the same as for regular annotation jobs
3. There was a function in UI that follows the current logic of the server to obtain correct frame meta

   cvat/cvat-ui/src/components/quality-control/task-quality/allocation-table.tsx

   Line 43 in 45e1b4b

   export function getAllocationTableContents(gtJobMeta: FramesMetaData, gtJob: Job): TableRowData[] {

   . A simplified version can be found here: Fix frame names in the allocation table for simple GT job #8731
4. Apparently, some recent (since mid December'24) fixes and refactorings in UI missed that difference in the endpoint output. The server API didn't change in this regard and works as before

I'd be great to support both variants of meta output for simple GT jobs. However, it requires changing server API. Specifically, it requires changing the "frames" field to return only frames from the "included_frames" range, if specified. I think we could add a request header or a parameter equivalent to ?return_placeholders=true/false for the /api/tasks/<id>/meta, then and migrate the UI to the updated version and gradually deprecate the old version.