Error when uploading MOT annotations

[yasakova-anastasia]

My actions before raising this issue

• Read/searched the docs
• Searched past issues

Expected Behaviour

Upload MOT annotations normally.

Current Behaviour

I have MOT annotations saved in .zip format with the following structure

$unzip -l tracking_results.zip 
Archive:  tracking_results.zip
  Length      Date    Time    Name
---------  ---------- -----   ----
    24059  2022-08-13 13:01   gt.txt
        3  2022-08-13 13:01   labels.txt
---------                     -------
    24062                     2 files

Here are the first few lines of the gt.txt file

0,0,0.7899999618530273,304.20000553131104,98.29000234603882,95.31000256538391,1,1,1.0
1,0,3.7700001150369644,301.1899948120117,98.28000068664551,95.31000256538391,1,1,1.0
2,0,6.7500002682209015,298.1800079345703,98.28000068664551,95.31000256538391,1,1,1.0
3,0,9.719999879598618,295.16000747680664,98.29000234603882,95.31999826431274,1,1,1.0
4,0,12.700000405311584,292.1499967575073,98.28000068664551,95.31000256538391,1,1,1.0
5,0,15.680000185966492,289.1400098800659,98.28000068664551,95.31000256538391,1,1,1.0
6,0,18.649999797344208,286.1299991607666,98.29000234603882,95.31000256538391,1,1,1.0
7,0,21.629999577999115,283.1199884414673,98.28000068664551,95.31000256538391,1,1,1.0
8,0,24.609999358654022,280.1000118255615,98.28000068664551,95.31999826431274,1,1,1.0
9,0,27.57999897003174,277.0900011062622,98.29000234603882,95.31000256538391,1,1,1.0

and the labels.txt file is simply

ok

However, when I try to upload these annotations using Upload annotations -> MOT 1.1 (on cvat.org I get an error saying

Error: Request failed with status code 500. "datumaro.components.errors.DatasetNotFoundError: Failed to find dataset at '/tmp/tmpcntvb5ty'\n".

What am I doing wrong? I think the format follows the one listed in the documentation.

Issue by @tadejsv

[sizov-kirill]

Issue successfully reproduced by me

The problem is that Datumaro looks for annotation files in the gt directory, but if the annotation files are stored in the root of the archive, it throws an error.

Another problem here is that provided annotation file uses incorrect frame numbering, according MOT specification frame numbers should start from 1

[zhiltsov-max]

@kirill-sizov , can we close the issue? Do you expect any fixes for this?

[sizov-kirill]

@zhiltsov-max Yes, I think we should support this use case, when annotations files are stored in the root of archive without gt sub-directory for MOT format

[niqbal996]

Hi @kirill-sizov,
I am facing a similar issue. I have created a a labelling job with ANY kind of labels. I am uploading the labels in COCO 1.1 format as a zip file. I have checked the label file locally with the official COCO API and the labels show up as expected. But CVAT throws an error saying:

cvat-app.tsx:295 Error: Request failed with status code 500. "Exception: Failed to find 'coco' dataset at '/tmp/tmpc61gumsx'\n"..

I do not think there is any fix posted here in this thread. I am using CVAT release version 2.2. Your help will be appreciated.
Thanks a lot.

[zhiltsov-max]

@niqbal996, if possible, please share the annotations you're uploading, so we could reproduce the problem. We don't need the images, but info about the file structure in the archive could be helpful (such as file tree).

[niqbal996]

Hi @zhiltsov-max ,
Here is the annotation file I am using
nirannotations.zip
Thanks for looking into this.

[rosebbb]

Hi @zhiltsov-max , Here is the annotation file I am using nirannotations.zip Thanks for looking into this.

Did you solve this? I have exactly the same problem with COCO format annotation file

[zhiltsov-max]

Hi, there is no fix yet. Please make sure your dataset file layout corresponds to the one described here.

[ixez]

It happened to me, but I figured out that I need to put gt.txt and labels.txt in a folder of gt inside the zip file.

[zhiltsov-max]

Related: #7052

For now, resolved by exclusion of the simplified format version from the documentation. However, such version can still be supported as it's simplifies the format usage, let's keep this issue as a feature request.