When importing labelled images in "Datumaro 1.0" format they are downloaded in point cloud format

[ganindu7]

Actions before raising this issue

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

Steps to Reproduce

when I try to import the dataset to cvat I get

"{
	"name": "MediaTypeError",
	"message": "Unexpected media type of a dataset item '000000'. Expected '<class 'datumaro.components.media.Image'>', actual '<class 'datumaro.components.media.PointCloudFromFile'>' ",
	"stack": "---------------------------------------------------------------------------

......
......

MediaTypeError: Unexpected media type of a dataset item '000000'. Expected '<class 'datumaro.components.media.Image'>', actual '<class 'datumaro.components.media.PointCloudFromFile'>' "
}

This is the same if I used the API or the Web GUI

Expected Behavior

the dataset be in "datumaro.components.media.Image" format. instead of Pointcloud

Possible Solution

Use the correct Datumaro version in CVAT or fix the bug.

Context

I updated to CAT 2.22.0 Yesterday but the issue still persists. not sure if the bit need for this needs to be updated separately or because the annotations were done before the update was made and was saved in the old format. I re exported few times after the update but in the downloaded annotation file I still get

...
[],"attr":{"frame":23},"point_cloud":{"path":""}},{"id":"000024","annotations":[]...
...

This issue is referenced #5924 but the solution is noted as updating CVAT (which I did) not sure if I'm doing it incorrectly or some residual issue.

Environment

No response

[zhiltsov-max]

Hi, it's not clear how you get such an error message, if you're uploading data into CVAT. There is no such name as PointCloudFromFile in the CVAT or Datumaro (our fork) codebases. If it's about the exported data, it was possible to get such a problem if annotations were exported without images before cvat-ai/datumaro#34 / CVAT 2.12.0. If you just upgraded to a version after 2.12.0, but exported the dataset before upgrading, try to change something in the task (e.g. name) and export one more time. CVAT caches exported datasets, so maybe this is causing the problem.

[ganindu7]

Hi Maxim, Thanks a lot for the reply!, Yes it is getting data out of CVAT (downloading annotations without images) I tried your tip. I renamed a previously unlabelled task, did some annotations and downloaded annotations ("Datumaro 1.0") format. this time I use the gui tools from the browser and the downloaded file still had point_cloud fields. my current cvat version is v2.22.0 (after the update) Server version: 2.22.0 Core version: 15.2.1 Canvas version: 2.20.10 UI version: 1.66.4 Cheers, Ganindu.

…

On Thu, 14 Nov 2024 at 12:50, Maxim Zhiltsov ***@***.***> wrote: Hi, it's not clear how you get such an error message, if you're uploading data into CVAT. There is no such name as PointCloudFromFile in the CVAT or Datumaro (our fork) codebases. If it's about the exported data, it was possible to get such a problem if annotations were exported without images before cvat-ai/datumaro#34 <cvat-ai/datumaro#34> / CVAT since 2.12.0 <70a7cc0>. If you just upgraded to a version after 2.12.0, but exported the dataset before upgrading, try to change something in the task (e.g. name) and export one more time. CVAT caches exported datasets, so maybe this causing the problem. — Reply to this email directly, view it on GitHub <#8700 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABG4JUT6JJTHOIIM4H3YRVT2ASMAXAVCNFSM6AAAAABRYWRQAGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINZWGI3TOMBTGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- *Best RegardsGanindu Nanayakkara*

[zhiltsov-max]

Hi, could you try to export the dataset with images?

[ganindu7]

Hi Max, This worked, I also tried downloading in "annotations-only" mode after downloading "with images" (that also gave the right type) then I labelled another previously unlabelled task. and downloaded only the annotations (no images) in "Datumaro 1.0" format and it didn't have point_cloud fields. does that make sense to you? if it does. I think we can mark this as resolved. and I guess this behaviour is still a borderline bug but I'll defer that judgement to you guys. Cheers, Ganindu.

…

On Fri, 15 Nov 2024 at 10:27, Maxim Zhiltsov ***@***.***> wrote: Hi, could you try to export the dataset with images? — Reply to this email directly, view it on GitHub <#8700 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABG4JUWN76JP3HYVPSFZP6T2AXD7PAVCNFSM6AAAAABRYWRQAGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINZYGUYDEOJUGY> . You are receiving this because you authored the thread.Message ID: ***@***.***>

-- *Best RegardsGanindu Nanayakkara*

[zhiltsov-max]

I also tried downloading in "annotations-only" mode after downloading "with images" (that also gave the right type)

This looks suspicious. Did the second file include only annotations?

I can see now there is actually a problem, which can be reproduced on the online platform. For this reason it makes sense to me to keep the issue open. There was another PR that was supposed to fix this, but it hasn't been merged yet.

[ganindu7]

Yes the second attempt was an annotation only attempt to see if the previous attempt with images have somehow changed something internal. Just to re-iterate in case of ambiguity The second “annotaion only” attempt following the successful “images with annotations” produced annotations that did not have a point_cloud field (which I wanted because that part of our system is automated to export to datumaro and merge with dataset after some other operations) So I will keep the issue open? Sent from Gmail Mobile

…

On Fri, 15 Nov 2024 at 11:42, Maxim Zhiltsov ***@***.***> wrote: I also tried downloading in "annotations-only" mode after downloading "with images" (that also gave the right type) This looks suspicious. Did the second file include only annotations? I can see now there is actually a problem, which can be reproduced on the online platform. For this reason it makes sense to me to keep the issue open. There was another PR <#7125> that was supposed to fix this, but it hasn't been merged yet. — Reply to this email directly, view it on GitHub <#8700 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/ABG4JUQ5L6YP7OTNOH7BCHL2AXM35AVCNFSM6AAAAABRYWRQAGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINZYGYZTINBQGQ> . You are receiving this because you authored the thread.Message ID: ***@***.***>

[zhiltsov-max]

Let's keep it until the PR mentioned is merged.