The ability to transform polygon shape to polygon track.

[medphisiker]

My actions before raising this issue

• [ ] I install CVAT on my Linux Ubuntu 20.04 server by this instruction link to docs. Then create a supervisor user and login. CVAT works perfectly.
• [ ] I create a task for labeling from the set of images.
• [ ] I read docs about "Track mode with polygons" link.
• [ ] Then I create track polygon on frame 1 (step 1) and press N key to finish it (step 2), moved to the frame 3 (step 3), press Shift+N to create the new polygon (step 4) and press N to finish (step 5).
• [ ] I read docs about CVAT AI Tools AI tools. I was very interested in Interactors for Semi-automatic segmentation like DEXTR, f-BRS, HRNet, Inside-Outside-Guidance.
• [ ] I follow guide in this page guide how install and run AI tools. I successfully launched DEXTR and it worked. For DEXTR running we have this commands:

nuctl deploy --project-name cvat \
  --path serverless/openvino/dextr/nuclio \
  --volume `pwd`/serverless/common:/opt/nuclio/common \
  --platform local

It is upload a special docker image with DEXTR, run it, and then CVAT can interact with it.

My idea and its description: I want to track some object, for example car1. I got polygon shape from DEXTR for car1 object on frame 1. Then I got polygon shape from DEXTR for car1 object on frame 7. I want to track car1 from frame 1 to frame 7. But as I understand I can't because DEXTR give me polygon shapes.

Some other intresting example. I have dataset with polygon shapes and I want to change it to tracking objects. It would be great to be able to simply combine polygons of the same object into one track. For example set to first polygon on frame1 you are belong to track1, then go to polygon on frame 2 set the same you are belong to track1 and ect.

Expected Behaviour

There is no bug in this case, there is a request to add additional functionality.

Current Behaviour

We can use integrators like DEXTR only for instance segmentation, but not for tracking. The dataset in which the data is labeled for instance segmentation, as I understand, cannot be converted to tracking.

Possible Solution

• Add the ability to convert first polygon shape to first key polygon track and convert second polygon shape to last key polygon track to make interpolation between them.
• Add the ability to combine different polygons on consecutive frames in one track.

Steps to Reproduce (for bugs)

There is no bug in this case, there is a request to add additional functionality.

Context

This will allow the use of interactors such as DEXTR for tracking objects. This will allow to combine different polygons on consecutive frames in one track. I think converting the dataset from instance segmentation(poly shapes) will be faster than labeling it from scratch from source images.

Your Environment

I have everything working according to the instructions. I'm just offering a new functionality.

[bsekachev]

Hi,

You can use "Merge" feature. On frame one enable merging mode and click the polygon shape. Go to frame seven and click the polygon there. Finish merging by clicking Merge button again. Go to frame two and remove outside keyframe there. (Outside keyframes are automatically added if there is not shape selected on frame N+1 after selected shape on frame N)

Now you get polygon track on frames 1-7. To correct interpolation set similar start point and direction using these buttons:

About merge you can find here: https://opencv.github.io/cvat/docs/manual/basics/track-mode-basics/

[medphisiker]

Thank you very much for fast answer. I try it and it works ) I also try with bounding boxes and and it works too. It may be good to add a small addition to the docs that combining tracks works with both tracking polygons and polygon shape (and with boxes too).

An interesting thing when exporting to MOTv1.1 I have tracks only from tracking bounding boxes. I thought that according to the existing tracking polygons, tracking bounding boxes would be created and they would be in the file gt.txt.

When I export it to COCO dataset I get bounding boxes automatically from tracking polygons (information about the tracks is lost because it detection dataset).

[bsekachev]

@zhiltsov-max

Do you have any suggestions about different behaviour in MOT and COCO?

[zhiltsov-max]

We had an idea of always trying to convert the annotations to the compatible form for the format, but we've decided that it would create more unexpected behavior than help. Currently, we minimize subtle annotation conversions only to the cases we sure it really makes sense to do always. COCO formats naturally support polygons and boxes simultaneously, that's why we use them there. MOT, in turn, is purely bbox tracks format, so no conversions are done.

Please consider using Datumaro for extra annotation conversions. For instance, the conversion from polygons to boxes can be done this way:

# > export to the Datumaro format, download & unzip first

pip install datumaro[default]
datum transform -t shapes_to_boxes -o output_dir extracted_dir/
datum convert -i output_dir/ -if datumaro -f mot_seq -o mot_dataset

information about the tracks is lost because it detection dataset

CVAT keeps the source track ids in the attributes section of the exported annotations.

[bsekachev]

Hope you got your answers. I will close the issue. Do not hesitate to reopen if you still have questions

[medphisiker]

Thank you for fast answering and help )

I tried to follow your advice and I get some following problems. I create a project and a annotation task for it from 10 images. I annotate ine tracking box and one tracking polygon to test export to MOT dataset. I go to my project and I export it as Datumaro dataset zip archive without images. I unzip it first to the same folder. I have a "annotations" folder with "default.json" file inside.

I work with anaconda python. To guarantee a clean experiment, I created a new working environment specifically for the package datumaro. I run command inside my python 3.10.8 virtual environment for datumaro:

pip install datumaro[default]

I have virtual environment with these modules:

# packages in environment at /home/admin-gpu/anaconda3/envs/datumaro:
#
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main  
_openmp_mutex             5.1                       1_gnu  
aiohttp                   3.8.3                    pypi_0    pypi
aiohttp-retry             2.8.3                    pypi_0    pypi
aiosignal                 1.3.1                    pypi_0    pypi
amqp                      5.1.1                    pypi_0    pypi
antlr4-python3-runtime    4.9.3                    pypi_0    pypi
appdirs                   1.4.4                    pypi_0    pypi
async-timeout             4.0.2                    pypi_0    pypi
asyncssh                  2.12.0                   pypi_0    pypi
atpublic                  3.1.1                    pypi_0    pypi
attrs                     22.1.0                   pypi_0    pypi
billiard                  3.6.4.0                  pypi_0    pypi
bzip2                     1.0.8                h7b6447c_0  
ca-certificates           2022.10.11           h06a4308_0  
celery                    5.2.7                    pypi_0    pypi
certifi                   2022.9.24       py310h06a4308_0  
cffi                      1.15.1                   pypi_0    pypi
charset-normalizer        2.1.1                    pypi_0    pypi
click                     8.1.3                    pypi_0    pypi
click-didyoumean          0.3.0                    pypi_0    pypi
click-plugins             1.1.1                    pypi_0    pypi
click-repl                0.2.0                    pypi_0    pypi
colorama                  0.4.6                    pypi_0    pypi
commonmark                0.9.1                    pypi_0    pypi
configobj                 5.0.6                    pypi_0    pypi
contourpy                 1.0.6                    pypi_0    pypi
cryptography              38.0.4                   pypi_0    pypi
cycler                    0.11.0                   pypi_0    pypi
datumaro                  0.3.1                    pypi_0    pypi
defusedxml                0.7.1                    pypi_0    pypi
dictdiffer                0.9.0                    pypi_0    pypi
diskcache                 5.4.0                    pypi_0    pypi
distro                    1.8.0                    pypi_0    pypi
dpath                     2.1.1                    pypi_0    pypi
dulwich                   0.20.50                  pypi_0    pypi
dvc                       2.36.0                   pypi_0    pypi
dvc-data                  0.28.3                   pypi_0    pypi
dvc-http                  2.27.2                   pypi_0    pypi
dvc-objects               0.14.0                   pypi_0    pypi
dvc-render                0.0.14                   pypi_0    pypi
dvc-task                  0.1.6                    pypi_0    pypi
dvclive                   1.1.0                    pypi_0    pypi
filelock                  3.8.0                    pypi_0    pypi
flatten-dict              0.4.2                    pypi_0    pypi
flufl-lock                7.1.1                    pypi_0    pypi
fonttools                 4.38.0                   pypi_0    pypi
frozenlist                1.3.3                    pypi_0    pypi
fsspec                    2022.11.0                pypi_0    pypi
funcy                     1.17                     pypi_0    pypi
future                    0.18.2                   pypi_0    pypi
gitdb                     4.0.10                   pypi_0    pypi
gitpython                 3.1.29                   pypi_0    pypi
grandalf                  0.6                      pypi_0    pypi
h5py                      3.7.0                    pypi_0    pypi
hydra-core                1.2.0                    pypi_0    pypi
idna                      3.4                      pypi_0    pypi
iterative-telemetry       0.0.6                    pypi_0    pypi
kiwisolver                1.4.4                    pypi_0    pypi
kombu                     5.2.4                    pypi_0    pypi
ld_impl_linux-64          2.38                 h1181459_1  
libffi                    3.4.2                h6a678d5_6  
libgcc-ng                 11.2.0               h1234567_1  
libgomp                   11.2.0               h1234567_1  
libstdcxx-ng              11.2.0               h1234567_1  
libuuid                   1.41.5               h5eee18b_0  
lxml                      4.9.1                    pypi_0    pypi
matplotlib                3.6.2                    pypi_0    pypi
multidict                 6.0.3                    pypi_0    pypi
nanotime                  0.5.2                    pypi_0    pypi
ncurses                   6.3                  h5eee18b_3  
networkx                  2.8.8                    pypi_0    pypi
nibabel                   4.0.2                    pypi_0    pypi
numpy                     1.23.5                   pypi_0    pypi
omegaconf                 2.2.3                    pypi_0    pypi
opencv-python             4.6.0.66                 pypi_0    pypi
openssl                   1.1.1s               h7f8727e_0  
orjson                    3.8.3                    pypi_0    pypi
packaging                 21.3                     pypi_0    pypi
pandas                    1.5.2                    pypi_0    pypi
pathspec                  0.9.0                    pypi_0    pypi
pillow                    9.3.0                    pypi_0    pypi
pip                       22.2.2          py310h06a4308_0  
prompt-toolkit            3.0.33                   pypi_0    pypi
protobuf                  3.20.1                   pypi_0    pypi
psutil                    5.9.4                    pypi_0    pypi
pycocotools               2.0.6                    pypi_0    pypi
pycparser                 2.21                     pypi_0    pypi
pydot                     1.4.2                    pypi_0    pypi
pygit2                    1.11.1                   pypi_0    pypi
pygments                  2.13.0                   pypi_0    pypi
pygtrie                   2.5.0                    pypi_0    pypi
pyparsing                 3.0.9                    pypi_0    pypi
python                    3.10.8               h7a1cb2a_1  
python-dateutil           2.8.2                    pypi_0    pypi
pytz                      2022.6                   pypi_0    pypi
pyyaml                    6.0                      pypi_0    pypi
readline                  8.2                  h5eee18b_0  
requests                  2.28.1                   pypi_0    pypi
rich                      12.6.0                   pypi_0    pypi
ruamel-yaml               0.17.21                  pypi_0    pypi
ruamel-yaml-clib          0.2.7                    pypi_0    pypi
scipy                     1.9.3                    pypi_0    pypi
scmrepo                   0.1.4                    pypi_0    pypi
setuptools                65.5.0          py310h06a4308_0  
shortuuid                 1.0.11                   pypi_0    pypi
shtab                     1.5.8                    pypi_0    pypi
six                       1.16.0                   pypi_0    pypi
smmap                     5.0.0                    pypi_0    pypi
sqlite                    3.40.0               h5082296_0  
tabulate                  0.9.0                    pypi_0    pypi
tensorboardx              2.5.1                    pypi_0    pypi
tk                        8.6.12               h1ccaba5_0  
tomlkit                   0.11.6                   pypi_0    pypi
tqdm                      4.64.1                   pypi_0    pypi
typing-extensions         4.4.0                    pypi_0    pypi
tzdata                    2022f                h04d1e81_0  
urllib3                   1.26.13                  pypi_0    pypi
vine                      5.0.0                    pypi_0    pypi
voluptuous                0.13.1                   pypi_0    pypi
wcwidth                   0.2.5                    pypi_0    pypi
wheel                     0.37.1             pyhd3eb1b0_0  
xz                        5.2.8                h5eee18b_0  
yarl                      1.8.2                    pypi_0    pypi
zc-lockfile               2.0                      pypi_0    pypi
zlib                      1.2.13               h5eee18b_0  

Then I run command

datum transform -t shapes_to_boxes -o output_dir annotations
# or datum transform -t shapes_to_boxes -o output_dir annotations/

And I get this error

$ datum transform -t shapes_to_boxes -o output_dir annotations/

ERROR:root:'NoneType' object has no attribute 'commit'
Traceback (most recent call last):
  File "/home/admin-gpu/anaconda3/envs/datumaro/bin/datum", line 8, in <module>
    sys.exit(main())
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/cli/__main__.py", line 184, in main
    retcode = args.command(args)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/util/scope.py", line 158, in wrapped_func
    ret_val = func(*args, **kwargs)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/cli/contexts/project/__init__.py", line 645, in transform_command
    dataset, _project = parse_full_revpath(args.target, project)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/cli/util/project.py", line 136, in parse_full_revpath
    return parse_revspec(s, ctx_project=ctx_project)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/util/scope.py", line 158, in wrapped_func
    ret_val = func(*args, **kwargs)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/cli/util/project.py", line 100, in parse_revspec
    if project.is_ref(proj_path):
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/project.py", line 2555, in is_ref
    return self._git.is_ref(ref)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/project.py", line 1174, in is_ref
    self.repo.commit(rev)
AttributeError: 'NoneType' object has no attribute 'commit'

For what reason does it occur? =)

[zhiltsov-max]

Seems strange, could you please check with the previous version of Datumaro (v0.3)? You can install it with pip install 'datumaro==0.3'. If not works, it deserves a bug report to the Datumaro repo. In both cases, you can use the following python script as a solution:

import datumaro as dm

dataset = dm.Dataset.import_from('annotations/', 'datumaro')
dataset.transform('shapes_to_boxes')
dataset.export('output_dataset', 'mot_seq') # add ", save_images=True" if you exported with images from CVAT

[medphisiker]

There seems to be some kind of error either in saving the file or in its path. For some reason, the script that you have given gives out such a strange error in my environment with datumaro == 0.3.1.

Traceback (most recent call last):
  File "/home/admin-gpu/Downloads/cvat_datasets/datumaro_convert.py", line 3, in <module>
    dataset = dm.Dataset.import_from('annotations/', 'datumaro')
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/dataset.py", line 1165, in import_from
    detected_sources = importer(path, **kwargs)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/extractor.py", line 440, in __call__
    raise DatasetNotFoundError(path)
datumaro.components.errors.DatasetNotFoundError: Failed to find dataset at 'annotations/'

I have tried both relative and absolute paths. Seems strange. Just in case, I attached a file "annotations/default.json" to this message (default.zip).

Just in case, I saved my test dataset to datumaro format again and extracted it. But nothing has changed.

Now I will create a new virtual environment with 'datumaro==0.3' and I'll try both ways with it.

[zhiltsov-max]

Ah, sorry. The path to the dataset should either be a path to the json file, or it should be a directory containing the annotations/ directory. Please try dataset = dm.Dataset.import_from('annotations/default.json', 'datumaro') or dataset = dm.Dataset.import_from("<the parent dir of annotations/>', 'datumaro').

[medphisiker]

I create new virtual environment and install datumaro==0.3. When running in console mode, I was asked to install git python, then the same error returned as before. I also ran the python script and got the same error as before. In general, the behavior of version 3.0 repeats the behavior of the new version 3.1. Python скрипт tells that there is not dataset at 'annotations/'. Perhaps for the same reason, when reading a folder 'annotations/' through the console, the code gets a NoneType and it does not have the necessary'commit' method that should be used further.

There is a suspicion that I seem to be getting the wrong zip archive of the datumaro dataset. And can I export with the same option of mask to shapes conversion from the cvat video data set format?

Below I give a list of installed packages in a new virtual environment with python 3.10 and datumaro 3.0.

# packages in environment at /home/admin-gpu/anaconda3/envs/datumaro:
#
# Name                    Version                   Build  Channel
_libgcc_mutex             0.1                        main  
_openmp_mutex             5.1                       1_gnu  
aiohttp                   3.8.3                    pypi_0    pypi
aiohttp-retry             2.8.3                    pypi_0    pypi
aiosignal                 1.3.1                    pypi_0    pypi
amqp                      5.1.1                    pypi_0    pypi
antlr4-python3-runtime    4.9.3                    pypi_0    pypi
appdirs                   1.4.4                    pypi_0    pypi
async-timeout             4.0.2                    pypi_0    pypi
asyncssh                  2.12.0                   pypi_0    pypi
atpublic                  3.1.1                    pypi_0    pypi
attrs                     22.1.0                   pypi_0    pypi
billiard                  3.6.4.0                  pypi_0    pypi
bzip2                     1.0.8                h7b6447c_0  
ca-certificates           2022.10.11           h06a4308_0  
celery                    5.2.7                    pypi_0    pypi
certifi                   2022.9.24       py310h06a4308_0  
cffi                      1.15.1                   pypi_0    pypi
charset-normalizer        2.1.1                    pypi_0    pypi
click                     8.1.3                    pypi_0    pypi
click-didyoumean          0.3.0                    pypi_0    pypi
click-plugins             1.1.1                    pypi_0    pypi
click-repl                0.2.0                    pypi_0    pypi
colorama                  0.4.6                    pypi_0    pypi
commonmark                0.9.1                    pypi_0    pypi
configobj                 5.0.6                    pypi_0    pypi
contourpy                 1.0.6                    pypi_0    pypi
cryptography              38.0.4                   pypi_0    pypi
cycler                    0.11.0                   pypi_0    pypi
datumaro                  0.3                      pypi_0    pypi
defusedxml                0.7.1                    pypi_0    pypi
dictdiffer                0.9.0                    pypi_0    pypi
diskcache                 5.4.0                    pypi_0    pypi
distro                    1.8.0                    pypi_0    pypi
dpath                     2.1.2                    pypi_0    pypi
dulwich                   0.20.50                  pypi_0    pypi
dvc                       2.36.0                   pypi_0    pypi
dvc-data                  0.28.3                   pypi_0    pypi
dvc-http                  2.27.2                   pypi_0    pypi
dvc-objects               0.14.0                   pypi_0    pypi
dvc-render                0.0.14                   pypi_0    pypi
dvc-task                  0.1.6                    pypi_0    pypi
dvclive                   1.1.0                    pypi_0    pypi
filelock                  3.8.1                    pypi_0    pypi
flatten-dict              0.4.2                    pypi_0    pypi
flufl-lock                7.1.1                    pypi_0    pypi
fonttools                 4.38.0                   pypi_0    pypi
frozenlist                1.3.3                    pypi_0    pypi
fsspec                    2022.11.0                pypi_0    pypi
funcy                     1.17                     pypi_0    pypi
future                    0.18.2                   pypi_0    pypi
gitdb                     4.0.10                   pypi_0    pypi
gitpython                 3.1.29                   pypi_0    pypi
grandalf                  0.6                      pypi_0    pypi
hydra-core                1.2.0                    pypi_0    pypi
idna                      3.4                      pypi_0    pypi
iterative-telemetry       0.0.6                    pypi_0    pypi
kiwisolver                1.4.4                    pypi_0    pypi
kombu                     5.2.4                    pypi_0    pypi
ld_impl_linux-64          2.38                 h1181459_1  
libffi                    3.4.2                h6a678d5_6  
libgcc-ng                 11.2.0               h1234567_1  
libgomp                   11.2.0               h1234567_1  
libstdcxx-ng              11.2.0               h1234567_1  
libuuid                   1.41.5               h5eee18b_0  
lxml                      4.9.1                    pypi_0    pypi
matplotlib                3.6.2                    pypi_0    pypi
multidict                 6.0.3                    pypi_0    pypi
nanotime                  0.5.2                    pypi_0    pypi
ncurses                   6.3                  h5eee18b_3  
networkx                  2.8.8                    pypi_0    pypi
numpy                     1.23.5                   pypi_0    pypi
omegaconf                 2.2.3                    pypi_0    pypi
opencv-python             4.6.0.66                 pypi_0    pypi
openssl                   1.1.1s               h7f8727e_0  
orjson                    3.8.3                    pypi_0    pypi
packaging                 21.3                     pypi_0    pypi
pandas                    1.5.2                    pypi_0    pypi
pathspec                  0.9.0                    pypi_0    pypi
pillow                    9.3.0                    pypi_0    pypi
pip                       22.2.2          py310h06a4308_0  
prompt-toolkit            3.0.33                   pypi_0    pypi
protobuf                  3.20.1                   pypi_0    pypi
psutil                    5.9.4                    pypi_0    pypi
pycocotools               2.0.6                    pypi_0    pypi
pycparser                 2.21                     pypi_0    pypi
pydot                     1.4.2                    pypi_0    pypi
pygit2                    1.11.1                   pypi_0    pypi
pygments                  2.13.0                   pypi_0    pypi
pygtrie                   2.5.0                    pypi_0    pypi
pyparsing                 3.0.9                    pypi_0    pypi
python                    3.10.8               h7a1cb2a_1  
python-dateutil           2.8.2                    pypi_0    pypi
pytz                      2022.6                   pypi_0    pypi
pyyaml                    6.0                      pypi_0    pypi
readline                  8.2                  h5eee18b_0  
requests                  2.28.1                   pypi_0    pypi
rich                      12.6.0                   pypi_0    pypi
ruamel-yaml               0.17.21                  pypi_0    pypi
ruamel-yaml-clib          0.2.7                    pypi_0    pypi
scipy                     1.9.3                    pypi_0    pypi
scmrepo                   0.1.4                    pypi_0    pypi
setuptools                65.5.0          py310h06a4308_0  
shortuuid                 1.0.11                   pypi_0    pypi
shtab                     1.5.8                    pypi_0    pypi
six                       1.16.0                   pypi_0    pypi
smmap                     5.0.0                    pypi_0    pypi
sqlite                    3.40.0               h5082296_0  
tabulate                  0.9.0                    pypi_0    pypi
tensorboardx              2.5.1                    pypi_0    pypi
tk                        8.6.12               h1ccaba5_0  
tomlkit                   0.11.6                   pypi_0    pypi
tqdm                      4.64.1                   pypi_0    pypi
typing-extensions         4.4.0                    pypi_0    pypi
tzdata                    2022f                h04d1e81_0  
urllib3                   1.26.13                  pypi_0    pypi
vine                      5.0.0                    pypi_0    pypi
voluptuous                0.13.1                   pypi_0    pypi
wcwidth                   0.2.5                    pypi_0    pypi
wheel                     0.37.1             pyhd3eb1b0_0  
xz                        5.2.8                h5eee18b_0  
yarl                      1.8.2                    pypi_0    pypi
zc-lockfile               2.0                      pypi_0    pypi
zlib                      1.2.13               h5eee18b_0  

[medphisiker]

I understand =) I tested this code and it didn't run I get new error. As I understand there is no mot_seq key word for MOT sequence dataset format. I went to read the documentation link but I did not find a description of keywords corresponding to the formats of datasets. On this page link MOT challenge dataset have name as you noted above 'mot_seq'.

Traceback (most recent call last):
  File "/home/admin-gpu/Downloads/cvat_datasets/datumaro_convert.py", line 6, in <module>
    dataset.export('output_dataset', 'mot_seq')
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/util/scope.py", line 135, in wrapped_func
    ret_val = func(*args, **kwargs)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/dataset.py", line 914, in export
    converter = self.env.converters[format]
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/environment.py", line 37, in __getitem__
    return self.get(key)
  File "/home/admin-gpu/anaconda3/envs/datumaro/lib/python3.10/site-packages/datumaro/components/environment.py", line 34, in get
    return self.items[key]
KeyError: 'mot_seq'

Perhaps this keyword 'mot_seq' for MOT challenge dataset appeared in version 3.1? Update: Test with datumaro==3.1, everything is the same as in version 3.0, the same error with the keyword 'mot_seq'.

[zhiltsov-max]

Ok, I've tested the pipeline. It seems, there are several problems along the path, which need to be fixed. A quick workaround will be to modify the script above this way:

import datumaro as dm

dataset = dm.Dataset.import_from('path/to/default.json', 'datumaro')
dataset.transform('shapes_to_boxes')
dataset.init_cache()
dataset._data._media_type = dm.Image
dataset.export('output_dataset', 'mot_seq_gt')

Here is the output_dataset.zip that I've got.

[medphisiker]

Ok, I've tested the pipeline. It seems, there are several problems along the path, which need to be fixed. A quick workaround will be to modify the script above this way:

import datumaro as dm

dataset = dm.Dataset.import_from('path/to/default.json', 'datumaro')
dataset.transform('shapes_to_boxes')
dataset.init_cache()
dataset._data._media_type = dm.Image
dataset.export('output_dataset', 'mot_seq_gt')

Here is the output_dataset.zip that I've got.

Thank you very much, I also test it and it works =) I also wanted to ask where to see which keywords correspond to the names of datasets? How to find that keyword for intresting dataset ? (MOT challenge is ''mot_seq_gt'') ? I search for MOT challenge dataset on documentation page link, but there is not MOT. But I find some other and find that for KITTI dataset we have keyword 'kitty' COCO dataset we have keyword 'coco'.

[zhiltsov-max]

How to find that keyword for interesting dataset ? (MOT challenge is ''mot_seq_gt'') ?

For many datasets there is documentation at link you attached, but not for all. For others, we can check in the format tests and in the full lists available in the datum import --help and datum export --help CLI output. Dataset examples are available in the test resources here, they should have their names aligned, for the most part.

[medphisiker]

datum import --help

Thank you for information and your help ) I have no more questions on this topic.

[gwengusc]

Hi,

You can use "Merge" feature. On frame one enable merging mode and click the polygon shape. Go to frame seven and click the polygon there. Finish merging by clicking Merge button again. Go to frame two and remove outside keyframe there. (Outside keyframes are automatically added if there is not shape selected on frame N+1 after selected shape on frame N)

Now you get polygon track on frames 1-7. To correct interpolation set similar start point and direction using these buttons:

About merge you can find here: https://opencv.github.io/cvat/docs/manual/basics/track-mode-basics/

Is there CVAT API for doing this?

[zhiltsov-max]

@gwengusc , it's a UI feature. Server API allows to download annotations and change them.