Dataset Management Framework, a Python library and a CLI tool to build, analyze and manage Computer Vision datasets.
Home Page: https://openvinotoolkit.github.io/datumaro/
License: MIT License
![dependabot[bot] avatar](https://avatars.githubusercontent.com/in/29110?v=4 "dependabot[bot]")
I'm missing a single class each time I annotate an image with polygons (enabled z order).
Environment used is via docker.
Scenario:
4 classes: a, b, c, d.
a - top left, b - top right, c - bottom right, d - bottom left (the annotation was done using this order).
I've exported the data via Segmentation mask 1.1.
SegmentationObject png:
But, every time a class is missing in SegmentationClass png:
When a dataset is being imported, its source files can contain errors in annotations. Currently, importing will fail, but there are other strategies possible. For example, an error can be ignored or fixed automatically. This would make dataset import more tolerant, allowing to partially import corrupted data.
FDDB (http://vis-www.cs.umass.edu/fddb/) is a famous dataset for face detection.
It uses an ellipse annotation.
How about adding the ellipse annotation type and supporting this dataset?
As I know, the ellipse annotation needs to be converted to BBox for training detection networks.
So I think it'd be good to support this functionality, too.
Currently, to add a new format, an Extractor plugin is needed, which can be accompanied by an Importer. In the essence, these 2 entities serve a single purpose - importing a dataset, so it looks like only one of them should be required for implementation.
An importer does:
Extractor(-s)An extractor does:
Hello :
I have two problems when I use upload function to upload labelme-format xml with text attributes.
First, if ',' in a sentence, the part after ',' will lost. For example, ABC CO.,LTD will lost ,LTD.
Second, numeric string will be auto processed to other string which represents a float number. For example "00001" will be transformed into "1.0" , "100.000" will be transformed into"100.00" .
Looking forward to your reply, thanks!
Requirements:
A video can contain lots of frames, most of which do not bring any new information. A dataset created from a video can be significantly reduced without affecting model performance, if we keep only the most significant frames.
References:
Command-line interface is one of the 2 main interfaces for the project. Currently, the only tested part of Datumaro - is its model, the library API, along with most of the plugins. CLI layer is being kept relatively shallow (according to the design document), so it usually simple to fix errors. But this puts the project in an uncomfortable state, because we don't really know if something is broken.
Introduction of such tests can also help to find mistakes in CLI design and make it simpler and better.
Requirements:
These things are not necessarily harmful in a CLI tool use cases, and they can even be desired in such scenario. However, they can be undesirable in a server-side use of the library.
chroot() call.I tried model run using openvino's mobileenet-v2-pytorch model.
(using mobilenet-v2-pytorch.xml, mobilenet-v2-pytorch.bin)
datum model run -p proj -m model-0
However, only the name of the layer (ex. 'prob' string) comes into the input parameters(outputs) of the interpreter. Please check the return result of OpenvinoLauncher.infer
results = self._net.infer(inputs) line 178, openvino_launcher.py
Debugging results are normal up to the code above, but it seems that only the name of the result layer is returned when returning and passing to interpreter.
We need at least one dataset for the person re-id task to utilize the task-specific split.
Market1501(https://deepai.org/dataset/market-1501) is a major dataset for the task.
I annotated two sets of images separately in CVAT and then want to combine those annotation files.
When I tried to combine two COCO JSON file annotated in CVAT, the annotations are not loading for the images.
When I load two separate annotation files then the annotation should be loaded for the images
Only few annotations are loading
git log -1):docker version (e.g. Docker 17.0.05):You may join our Gitter channel for community support.
Vision datasets often include 3d data - it can be lidar, MRI, landscape, buildings and models. It would be nice is Datumaro could handle such datasets. Examples: BRATS, KITTI, ObjectNet3D, MVTec ITODD, T-LESS, Falling Things
In CVAT it is necessary to workaround Datumaro installation. See the PR for details: cvat-ai/cvat#2842. Could you please investigate and provide an appropriate solution for the issue?
# workaround for binary incompatibility with numpy when pycocotools is installed by wheel
datumaro==0.1.5.1 --no-binary=datumaro --no-binary=pycocotoolsINFO: pip is looking at multiple versions of click to determine which version is compatible with other requirements. This could take a while.
INFO: pip is looking at multiple versions of av to determine which version is compatible with other requirements. This could take a while.
ERROR: Cannot install datumaro==0.1.5.1, h5py==2.10.0, numpy==1.20.0, open3d==0.11.2, opencv-python-headless==4.4.0.42 and tensorflow==2.4.1 because these package versions have conflicting dependencies.
The conflict is caused by:
The user requested numpy==1.20.0
datumaro 0.1.5.1 depends on numpy>=1.17.3
h5py 2.10.0 depends on numpy>=1.7
open3d 0.11.2 depends on numpy
opencv-python-headless 4.4.0.42 depends on numpy>=1.17.3
tensorflow 2.4.1 depends on numpy~=1.19.2
To fix this you could try to:
1. loosen the range of package versions you've specified
2. remove package versions to allow pip attempt to solve the dependency conflictWhen there is a dataset, it is often important to be able to check its quality. There are few way to check it:
It would be nice to support these and other approaches to quality checking.
Currently, in 0.1.6, the approaches 2 and 3 are implemented as datum merge.
I tried to view the CVAT project state exported using "datumaro 1.0" and got the following issue.
Below are steps I took to view the project stats:
cd "../task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0"
datumaro project stats
OS: Ubuntu 18.04
Installation: pip install 'git+https://github.com/openvinotoolkit/datumaro'
Datumaro version (datumaru --version) : 0.1.0
Any advice would be appreciated !
2020-09-28 14:31:35,279 ERROR: HTTPConnectionPool(host='http', port=80): Max retries exceeded with url: //192.168.1.43:8080/api/v1/auth/login (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7effe13189a0>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution'))
Traceback (most recent call last):
File "/usr/lib/python3/dist-packages/urllib3/connection.py", line 156, in _new_conn
conn = connection.create_connection(
File "/usr/lib/python3/dist-packages/urllib3/util/connection.py", line 61, in create_connection
for res in socket.getaddrinfo(host, port, family, socket.SOCK_STREAM):
File "/usr/lib/python3.8/socket.py", line 918, in getaddrinfo
for res in _socket.getaddrinfo(host, port, family, type, proto, flags):
socket.gaierror: [Errno -3] Temporary failure in name resolution
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/usr/lib/python3/dist-packages/urllib3/connectionpool.py", line 665, in urlopen
httplib_response = self._make_request(
File "/usr/lib/python3/dist-packages/urllib3/connectionpool.py", line 387, in _make_request
conn.request(method, url, **httplib_request_kw)
File "/usr/lib/python3.8/http/client.py", line 1240, in request
self._send_request(method, url, body, headers, encode_chunked)
File "/usr/lib/python3.8/http/client.py", line 1286, in _send_request
self.endheaders(body, encode_chunked=encode_chunked)
File "/usr/lib/python3.8/http/client.py", line 1235, in endheaders
self._send_output(message_body, encode_chunked=encode_chunked)
File "/usr/lib/python3.8/http/client.py", line 1006, in _send_output
self.send(msg)
File "/usr/lib/python3.8/http/client.py", line 946, in send
self.connect()
File "/usr/lib/python3/dist-packages/urllib3/connection.py", line 184, in connect
conn = self._new_conn()
File "/usr/lib/python3/dist-packages/urllib3/connection.py", line 168, in _new_conn
raise NewConnectionError(
urllib3.exceptions.NewConnectionError: <urllib3.connection.HTTPConnection object at 0x7effe13189a0>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/usr/lib/python3/dist-packages/requests/adapters.py", line 439, in send
resp = conn.urlopen(
File "/usr/lib/python3/dist-packages/urllib3/connectionpool.py", line 719, in urlopen
retries = retries.increment(
File "/usr/lib/python3/dist-packages/urllib3/util/retry.py", line 436, in increment
raise MaxRetryError(_pool, url, error or ResponseError(cause))
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='http', port=80): Max retries exceeded with url: //192.168.1.43:8080/api/v1/auth/login (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7effe13189a0>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution'))
During handling of the above exception, another exception occurred:
Traceback (most recent call last):
File "/usr/lib/python3.8/runpy.py", line 193, in _run_module_as_main
return _run_code(code, main_globals, None,
File "/usr/lib/python3.8/runpy.py", line 86, in _run_code
exec(code, run_globals)
File "/usr/local/lib/python3.8/dist-packages/datumaro/__main__.py", line 12, in <module>
sys.exit(main())
File "/usr/local/lib/python3.8/dist-packages/datumaro/cli/__main__.py", line 115, in main
return args.command(args)
File "/usr/local/lib/python3.8/dist-packages/datumaro/cli/contexts/project/__init__.py", line 725, in stats_command
stats.update(compute_image_statistics(dataset))
File "/usr/local/lib/python3.8/dist-packages/datumaro/components/operations.py", line 1051, in compute_image_statistics
stats['dataset'].update(_extractor_stats(dataset))
File "/usr/local/lib/python3.8/dist-packages/datumaro/components/operations.py", line 1039, in _extractor_stats
mean, std = mean_std(extractor)
File "/usr/local/lib/python3.8/dist-packages/datumaro/components/operations.py", line 962, in mean_std
image = item.image.data
File "/usr/local/lib/python3.8/dist-packages/datumaro/util/image.py", line 225, in data
return self._data()
File "/usr/local/lib/python3.8/dist-packages/datumaro/util/image.py", line 180, in __call__
image = self.loader(self.path)
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/.datumaro/plugins/cvat_rest_api_task_images.py", line 33, in <lambda>
lambda item_id: self._image_loader(item_id, self))
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/.datumaro/plugins/cvat_rest_api_task_images.py", line 62, in _image_loader
extractor._download_image(item_id)
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/.datumaro/plugins/cvat_rest_api_task_images.py", line 39, in _download_image
self._connect()
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/.datumaro/plugins/cvat_rest_api_task_images.py", line 56, in _connect
cli = CVAT_CLI(session, api, credentials=(username, password))
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/cvat/utils/cli/core/core.py", line 24, in __init__
self.login(credentials)
File "/home/canoo/Downloads/task_test-sever-20200929-2020_09_28_05_00_34-datumaro 1.0/cvat/utils/cli/core/core.py", line 175, in login
response = self.session.post(url, auth)
File "/usr/lib/python3/dist-packages/requests/sessions.py", line 581, in post
return self.request('POST', url, data=data, json=json, **kwargs)
File "/usr/lib/python3/dist-packages/requests/sessions.py", line 533, in request
resp = self.send(prep, **send_kwargs)
File "/usr/lib/python3/dist-packages/requests/sessions.py", line 646, in send
r = adapter.send(request, **kwargs)
File "/usr/lib/python3/dist-packages/requests/adapters.py", line 516, in send
raise ConnectionError(e, request=request)
requests.exceptions.ConnectionError: HTTPConnectionPool(host='http', port=80): Max retries exceeded with url: //192.168.1.43:8080/api/v1/auth/login (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7effe13189a0>: Failed to establish a new connection: [Errno -3] Temporary failure in name resolution'))
Datumaro supports many datasets, each of which has its own data layout and properties. When a dataset is exported, there can be lots of conditions and expectations to the input dataset. There are a number of import and export options in each format. Currently, the only documentation for them is:
It would be nice, if Datumaro had more comprehensive documentation for formats, import and export parameters, imported dataset look and exported dataset expectations.
A common problem for researchers is having large images with lots of small objects. Is there a function in datumaro to tile images (overlapping option possible, min_visibility option possible) into specified dimensions and the corresponding masks, bounding boxes, and polygons also get tiled, with images renamed with their original coordinates. If not, then it would be great to have this functionality.
Cheers, Sam
Need to document at least interfaces using docstrings. For example: Dataset, DatasetItem, etc.
For more information and reference: https://realpython.com/documenting-python-code/#documenting-your-python-projects
Proposed docstring fromat: https://numpydoc.readthedocs.io/en/latest/format.html
Once a dataset is located or imported, it can be useful to provide source file information and other format-related properties, such as image directory, source files, presence of images etc.
Similarly, when a dataset is exported, it can be useful to know, which files were created or updated.
Visual datasets are not restricted by images and often come from video frames. The ability to use video data is essential for such project as Datumaro.
Requirements:
Ø Guidance:
To export subsets,
Example: extract a dataset with only images of subset 'train'.
datum project filter \
-p test_project \
-e '/item[subset="train"]'
Ø Actual Result:
datum project create -o proj_splitter
datum source add path imagenet/val -f imagenet -p proj_splitter/
datum project transform -t classification_split -p proj_splitter/ -- --subset train:.5 --subset val:.2 --subset test:.3
datum project filter -p proj_splitter -e '/item[subset="train"]'
datum project filter -p proj_splitter -e '/item[subset="val"]'
datum project filter -p proj_splitter -e '/item[subset="test"]'
Export message like 'Subproject has been extracted to '/home/cvalgo/workspace/datumaro_splitter/datumaro/proj_splitter-filter' is shown, but the folder is empty.
Hi,
Could you please confirm if I take two coco datasets as input and output their merged dataset in coco format?
Would this library handle the datasets if both have images with same names?
Try to export segmentations masks according to the MOTS format (see attachment)
I'm trying to build a new Multiple object tracking and segmentation dataset. For this I'm eager to find a annotation tool that can provide me the right annotations formats used in the MOTS dataset (.png or .txt). Hence, supporting temporally
consistent object instance segmentations.
Clouds and other kinds of remotes, which include s3, Azure, Google Drive, NFS & CIFS file shares, Git, Git-lfs, etc. are common places for storing large datasets. Support of such storages can be very useful when working with datasets.
Requirements:
Datasets often come as zip or tar archives, it would be nice to support such kind of inputs in Datumaro. It can be undesirable to unpack them, because this can have a major impact on storage.
Requirements:
tar.gz, ((file.zip).zip).7zHi, how do I remove images without annotations?
I tried datum project filter -m i+a, but I don't know what to put for the -e param.
This might work datum project filter -m i+a -e 'item/annotation[label!="dswfyhsdyf8"]' .. but the images are not carried over to the subproject (empty directory).
Also, how do I pass a list of images to exclude from the project? (e.g half annotated images/poorly annotated images that I don't want to use for train/val)
VGGFace2 is basically FR (Face recognition) dataset,
but our implementation only supports landmark and bounding box annotations.
I think we should add Label(...) so it can be used for FR task.
And many fr data has the same structure except the landmark information.
That is, a directory represents a person and it contains images for the person.
So if we make landmark annotations to be optional, we can support lots of FR data.
In summary,
I am trying to install datumaro with:
pip install datumaroBut I get the following error message:
Collecting datumaro
Using cached datumaro-0.1.5.1-py3-none-any.whl (163 kB)
ERROR: Could not find a version that satisfies the requirement pycocotools-windows; platform_system == "Windows" (from datumaro)
ERROR: No matching distribution found for pycocotools-windows; platform_system == "Windows"I'm on Windows 10
Hi,
I have a requirement to import a 3rd party mask dataset into CVAT as polygon for review - The mask sample is attached
Can you advise which steps should I take to convert it into CVAT supported format (such as : segmenetation mask 1.1) using datumaro?
script
from datumaro.components.project import Environment
from datumaro.plugins.yolo_format.converter import YoloConverter
# Import and save a dataset
env = Environment()
dataset = env.make_importer("cvat")(str(self.dataset_path)).make_dataset()
env.converters.get("yolo").convert(dataset, save_dir="dst/dir")
Dataset folder structure
---dataset
|----report.xml
|----images
|-----forest1
|-----1.jpg
|-----2.jpg
|-----forest2
|---3.jpg
While converting CVAT into yolo format by using above code, got following error ,,
Traceback (most recent call last):
File "main.py", line 118, in
app.run(main)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/absl/app.py", line 300, in run
_run_main(main, args)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/absl/app.py", line 251, in _run_main
sys.exit(main(argv))
File "main.py", line 114, in main
ss.produce()
File "main.py", line 77, in produce
handle_dataset.cvat2yolo()
File "/home/sam/ss/utils/dataset_generator.py", line 135, in cvat2yolo
dataset = env.make_importer("cvat")(str(self.dataset_path)).make_dataset()
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/datumaro/components/project.py", line 600, in make_dataset
return ProjectDataset(self)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/datumaro/components/project.py", line 318, in init
sources[s_name] = env.make_extractor(s_format, url, **options)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/datumaro/components/project.py", line 284, in make_extractor
return self.extractors.get(name)(*args, **kwargs)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/datumaro/plugins/cvat_format/extractor.py", line 33, in init
items, categories = self._parse(path)
File "/root/anaconda3/envs/aug/lib/python3.8/site-packages/datumaro/plugins/cvat_format/extractor.py", line 94, in _parse
attr_value = float(attr_value)
TypeError: float() argument must be a string or a number, not 'NoneType'
Dumping Annotations in LabelMe3.0 Format. Annotations should be dumped with File (Image) name and should remove the base path (head of the file name).
Eg. root/Folder_containing_images/Image1.jpg -- > Annotations in LabelMe3.0 should be dumped as Image1.xml
Currently, I am uploading the image data from the shared drive on the CVAT tool. Hence the Images are uploaded with base path + filename (root/Folder_containing_images/Image1.jpg). While Dumping Annotations in LabelMe3.0, it raises an error as it is only expecting a flat layout for the Image name
Support Relative Path/ Absolute Path for LabelMe3.0.
Reference - https://github.com/openvinotoolkit/cvat/pull/1715/files
I am trying to Upload the Annotations file to verify and correct the Labels. LabelMe3.0 suits for the application as it generates an individual XML file for each image and supports polygon labels and attributes as well. Converters are designed to/from LabelMe3.0 to a custom format. It will be great if it is possible to support the relative/absolute path (without flat layout) in the LabelMe3.0 format as well.
Also, Open for suggestions if any other Annotations formats can be suitable (generates a separate file and supports polygon and attributes) and used.
Thanks a lot!
CVAT tool is deployed on AWS instance
A single PDF can be considered as a container with several images, each of which can be used for annotation. Look at CVAT implementation of PDF data as a reference.
A list of code and architecture improvements, refactorings etc.:
occluded as a class member for shapes and masks instead of a free attributeproperties along with attributes so attributes could save their specific meaningAnnotationType, just use typesgather singular points to skeleton transformDataset subdivisions, "virtual" subsets (non-owning views)CLI:
I have 12 4K images which I have annotated with polygons. When I try to download the annotation with the segmentation mask, it takes a long time. I am still waiting from more 1 hours. The website console seems normal and no errors are issued.
To my understanding, changing the points to a mask shall be very fast and should not take a long time.
The images are aerial images and they contain very small to big polygons.
Exporting annotations shouldn't fail because of image filename. Task creation and image import must failed with precise error.
It seems that I've go a task completly messed up by images with ".jpeg" extension in filename, (ex: 1e7ca95c572a8227070dc68badba1e18-79860.241744.jpeg).
If it's the case, a proper filename check test should prevent this error.
This action prevent people to silently break their collection.
We've got a worklow that generate image name with .jpeg extension, trying to export those file trigger error.
git log -1):docker version (e.g. Docker 17.0.05):cvat_proxy | 172.28.0.1 - - [06/Oct/2020:18:29:16 +0000] "GET /api/v1/tasks/5/dataset?format=TFRecord%201.0 HTTP/1.1" 202 0 "http://localhost:8080/tasks/5/jobs/5" "Mozilla/5.0 (X11; Linux x86_64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/85.0.4183.121 Safari/537.36"
cvat | 2020-10-06 18:29:16,972 DEBG 'rqworker_default_0' stderr output:
cvat | DEBUG:rq.worker:Sent heartbeat to prevent worker timeout. Next one should arrive within 480 seconds.
cvat |
cvat | 2020-10-06 18:29:16,979 DEBG 'rqworker_default_0' stderr output:
cvat | DEBUG:rq.worker:Sent heartbeat to prevent worker timeout. Next one should arrive within 90 seconds.
cvat |
cvat | 2020-10-06 18:29:17,064 DEBG 'rqworker_default_0' stderr output:
**cvat | WARNING:root:Item '540d5b824f715a53d994389147ceeb53-80219.032895': can't find format string for the '.jpeg' image extension, the corresponding field will be empty.**
cvat |
cvat | 2020-10-06 18:29:17,508 DEBG 'rqworker_default_0' stderr output:
cvat | [2020-10-06 18:29:17,507] ERROR cvat.server.task_5: [cvat.apps.dataset_manager.views @ export_task]: exception occurred
cvat | Traceback (most recent call last):
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 61, in export_task
cvat | task.export_task(task_id, temp_file, dst_format,
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 748, in export_task
cvat | task.export(f, exporter, host=server_url, save_images=save_images)
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 630, in export
cvat | exporter(dst_file, task_data, **options)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/registry.py", line 36, in __call__
cvat | f_or_cls(*args, **kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/tfrecord.py", line 30, in _export
cvat | dm_env.converters.get('tf_detection_api').convert(extractor,
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/components/converter.py", line 37, in convert
cvat | return converter.apply()
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 100, in apply
cvat | tf_example = self._make_tf_example(item)
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 194, in _make_tf_example
cvat | 'image/format': bytes_feature(fmt.encode('utf-8')),
cvat | AttributeError: 'NoneType' object has no attribute 'encode'
cvat |
cvat | 2020-10-06 18:29:17,508 DEBG 'rqworker_default_0' stderr output:
cvat | ERROR:cvat.server.task_5:[cvat.apps.dataset_manager.views @ export_task]: exception occurred
cvat | Traceback (most recent call last):
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 61, in export_task
cvat | task.export_task(task_id, temp_file, dst_format,
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 748, in export_task
cvat | task.export(f, exporter, host=server_url, save_images=save_images)
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 630, in export
cvat | exporter(dst_file, task_data, **options)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/registry.py", line 36, in __call__
cvat | f_or_cls(*args, **kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/tfrecord.py", line 30, in _export
cvat | dm_env.converters.get('tf_detection_api').convert(extractor,
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/components/converter.py", line 37, in convert
cvat | return converter.apply()
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 100, in apply
cvat | tf_example = self._make_tf_example(item)
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 194, in _make_tf_example
cvat | 'image/format': bytes_feature(fmt.encode('utf-8')),
cvat | AttributeError: 'NoneType' object has no attribute 'encode'
cvat |
cvat | 2020-10-06 18:29:17,508 DEBG 'rqworker_default_0' stderr output:
cvat | DEBUG:rq.worker:Handling failed execution of job /api/v1/tasks/5/dataset/TFRecord 1.0
cvat |
cvat | 2020-10-06 18:29:17,510 DEBG 'rqworker_default_0' stderr output:
cvat | ERROR:rq.worker:Traceback (most recent call last):
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/worker.py", line 936, in perform_job
cvat | rv = job.perform()
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/job.py", line 684, in perform
cvat | self._result = self._execute()
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/job.py", line 690, in _execute
cvat | return self.func(*self.args, **self.kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 84, in export_task_as_dataset
cvat | return export_task(task_id, dst_format, server_url=server_url, save_images=True)
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 61, in export_task
cvat | task.export_task(task_id, temp_file, dst_format,
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 748, in export_task
cvat | task.export(f, exporter, host=server_url, save_images=save_images)
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 630, in export
cvat | exporter(dst_file, task_data, **options)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/registry.py", line 36, in __call__
cvat | f_or_cls(*args, **kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/tfrecord.py", line 30, in _export
cvat | dm_env.converters.get('tf_detection_api').convert(extractor,
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/components/converter.py", line 37, in convert
cvat | return converter.apply()
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 100, in apply
cvat | tf_example = self._make_tf_example(item)
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 194, in _make_tf_example
cvat | 'image/format': bytes_feature(fmt.encode('utf-8')),
cvat | AttributeError: 'NoneType' object has no attribute 'encode'
cvat | Traceback (most recent call last):
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/worker.py", line 936, in perform_job
cvat | rv = job.perform()
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/job.py", line 684, in perform
cvat | self._result = self._execute()
cvat | File "/usr/local/lib/python3.8/dist-packages/rq/job.py", line 690, in _execute
cvat | return self.func(*self.args, **self.kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 84, in export_task_as_dataset
cvat | return export_task(task_id, dst_format, server_url=server_url, save_images=True)
cvat | File "/home/django/cvat/apps/dataset_manager/views.py", line 61, in export_task
cvat | task.export_task(task_id, temp_file, dst_format,
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 748, in export_task
cvat | task.export(f, exporter, host=server_url, save_images=save_images)
cvat | File "/home/django/cvat/apps/dataset_manager/task.py", line 630, in export
cvat | exporter(dst_file, task_data, **options)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/registry.py", line 36, in __call__
cvat | f_or_cls(*args, **kwargs)
cvat | File "/home/django/cvat/apps/dataset_manager/formats/tfrecord.py", line 30, in _export
cvat | dm_env.converters.get('tf_detection_api').convert(extractor,
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/components/converter.py", line 37, in convert
cvat | return converter.apply()
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 100, in apply
cvat | tf_example = self._make_tf_example(item)
cvat | File "/usr/local/lib/python3.8/dist-packages/datumaro/plugins/tf_detection_api_format/converter.py", line 194, in _make_tf_example
cvat | 'image/format': bytes_feature(fmt.encode('utf-8')),
cvat | AttributeError: 'NoneType' object has no attribute 'encode'
cvat |
cvat | 2020-10-06 18:29:17,511 DEBG 'rqworker_default_0' stderr output:
cvat | DEBUG:rq.worker:Invoking exception handler <function rq_handler at 0x7f9b36157e50>
You may join our Gitter channel for community support.
Hi,
From CVAT, I exported two tasks as datasets using the "datumaro 1.0" format such that I obtain two archives directory "task1.zip" and "task2.zip". My goal is to combine these two datasets into one so that I can feed the resulting dataset to the YOLO framework.
Here are the steps I take from there:
datum create -o projdatum add path -p proj -f datumaro task1
datum add path -p proj -f datumaro task2datum export -p proj -f yolo -- --save-imagesAs a result, a directory is created with what seems to be the files and format expected by the YOLO framework, but for some reason, it only contains part of the data, here the annotations and images from "task1".
I would expect the output directory to contain the complete dataset, which is here the annotations and images from "task1" and from "task2".
hi! i want to filter from data of cvat format to yolo. the source format is as follows:
....
<image id="0" name="frame_000000" width="2560" height="1440">
<box label="ball" occluded="0" source="manual" xtl="2129.51" ytl="854.22" xbr="2149.60" ybr="876.12">
<attribute name="target">true</attribute>
<attribute name="Fake">false</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2042.22" ytl="551.81" xbr="2057.12" ybr="567.73">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1798.30" ytl="404.61" xbr="1808.95" ybr="416.40">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1741.19" ytl="380.51" xbr="1750.72" ybr="391.15">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1733.48" ytl="373.91" xbr="1743.94" ybr="383.63">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1372.81" ytl="212.90" xbr="1380.73" ybr="221.08">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2190.33" ytl="503.11" xbr="2198.47" ybr="510.81">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2307.02" ytl="506.72" xbr="2314.06" ybr="513.98">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2256.42" ytl="477.24" xbr="2263.02" ybr="483.84">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1983.12" ytl="360.50" xbr="1990.90" ybr="368.50">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2020.33" ytl="357.68" xbr="2025.63" ybr="362.98">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="660.04" ytl="110.90" xbr="677.70" ybr="127.79">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1177.99" ytl="138.10" xbr="1196.80" ybr="157.38">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1197.60" ytl="139.40" xbr="1211.80" ybr="158.06">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1653.91" ytl="229.90" xbr="1673.00" ybr="247.79">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="811.55" ytl="98.90" xbr="829.90" ybr="115.94">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
</image>
<image id="1" name="frame_000001" width="2560" height="1440">
<box label="ball" occluded="0" source="manual" xtl="2134.86" ytl="848.87" xbr="2154.95" ybr="870.77">
<attribute name="target">true</attribute>
<attribute name="Fake">false</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2042.22" ytl="551.81" xbr="2057.12" ybr="567.73">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1798.30" ytl="404.61" xbr="1808.95" ybr="416.40">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1741.19" ytl="380.51" xbr="1750.72" ybr="391.15">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1733.48" ytl="373.91" xbr="1743.94" ybr="383.63">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1372.81" ytl="212.90" xbr="1380.73" ybr="221.08">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2193.02" ytl="500.42" xbr="2201.16" ybr="508.12">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2307.02" ytl="506.72" xbr="2314.06" ybr="513.98">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2256.42" ytl="477.24" xbr="2263.02" ybr="483.84">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1983.12" ytl="360.50" xbr="1990.90" ybr="368.50">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="2020.33" ytl="357.68" xbr="2025.63" ybr="362.98">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="660.04" ytl="110.90" xbr="677.70" ybr="127.79">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1177.99" ytl="138.10" xbr="1196.80" ybr="157.38">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1197.60" ytl="139.40" xbr="1211.80" ybr="158.06">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="1653.91" ytl="229.90" xbr="1673.00" ybr="247.79">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
<box label="ball" occluded="0" source="manual" xtl="811.55" ytl="98.90" xbr="829.90" ybr="115.94">
<attribute name="target">false</attribute>
<attribute name="Fake">true</attribute>
</box>
</image>
....
I just want to keep the data that satisfies non-occlusion and target is equal to true. How to write the command line?
thks!
Description:
RandomSplit transformation does not randomise entries in dataset and generates exactly the same output each time it is executed. What's more, it splits the dataset without mixing their indices, which may result in very unbalanced subsets.
Expected behaviour:
Each use of RandomSplit transformation should generate different subsets with mixed dataset entries.
Reproduction:
import os
import random
from datumaro.components.extractor import Image, DatasetItem, Label
from datumaro.components.project import Dataset, Environment
location = '/tmp/bug_reproduction'
env = Environment()
categories = ['dog', 'cat', 'frog', 'monkey', 'lion']
example = Dataset.from_iterable([
DatasetItem(
id=f'image{i}',
image=Image(path='abc.jpg', size=(32, 32)),
annotations=[Label(random.randint(0, len(categories)-1))],
) for i in range(100)
], categories=categories)
example = example.transform(env.transforms.get('random_split'), [('train', 0.5), ('val', 0.5)])
example = Dataset.from_extractors(example)
env.converters.get('voc').convert(example, location)
with open(os.path.join(location, 'ImageSets', 'Main', 'train.txt')) as train_txt:
print(train_txt.read())As a result, „train.txt” file contains images indexed from 0 up to 49 in a consecutive order:
image0
image1
image2
image3
...
image47
image48
image49
I took a look into Datumaro code and dataset indices are indeed shuffled. However, they are no longer used after that:
https://github.com/openvinotoolkit/datumaro/blob/develop/datumaro/plugins/transforms.py#L360
Imagenet images have various extensions, but datumaro only supports '.jpg'.
We should support more extensions likes in mmclassification (https://github.com/open-mmlab/mmclassification)
IMG_EXTENSIONS = ('.jpg', '.jpeg', '.png', '.ppm', '.bmp', '.pgm', '.tif')VOC annotations might contain the <difficult> tag inside of an object. This is seldom used however I had a pytorch project needing this. I used <difficult>0</difficult> in all object annotations then.
Further more the bandwidth depth of the images is not set correctly.
Hi there. I'm a big fan of using the openvino model assisted annotation. So I need to train a MaskRCNN and convert it to openvino then upload the model to cvat (have done it for faster rcnn). But to train MaskRCNN you need a TFrecord file and for that you need the XMLS( in the correct format) with the bounding box coordinates of the polygons and also the PNG masks. When I dump annotations in the sementation mask format I don't get the XMLS. When I export in other format coco it has a json file which has the bbox co-ords but that is not what's needed.. We need an xml per image not one big Json file in the wrong format.
CVAT doesn't export the write annotation files needed to (convert to tfrecord) and train maskRCNN.
I'm Running CVAT in docker. Using a slightly older version - just before serverless functions were introduced.
Cheers,
Sam
Datumaro comes as a command-line tool and as a Python library. While it has a command-line interface, it does not provide a stable an simple to use library API. Such API needs to be documented and tested.
Requirements:
import datumaro or import datumaro.apiDataset, annotation types, etc.)After releasing a dataset, it is often needs to be updated and maintained. In this process, or even in the process of experimenting and building the first version, it is very desirable to be able to reproduce any previous state of a dataset. Some solutions already provide similar functionality, so they might be helpful in implementation: DVC.
Requirements:
dataset length: 0
If an annotation in COCO format has category_id = 0, it will be considered as having no label information. However, categories can have any integer id value, including 0, so the real category could be ignored.
Original COCO annotations have indices starting at 1, so there is no problem.
Loading of plugins can be expensive, as it actually is - for TF and Accuracy Checker plugins, which can be loaded in few seconds. Plugins need to be reorganized in such way, that their loading is deferred until their invocation. The list of plugins, however, must be loaded quickly.
Currently, all annotations can have arbitrary attributes, but there is no specifications for these attributes. It would be great, if Datumaro provided:
Most dataset formats do not include such information, however. As the first step, we could:
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
Django
The Web framework for perfectionists with deadlines.
Laravel
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
Microsoft
Open source projects and samples from Microsoft.
Google
Google ❤️ Open Source for everyone.
Alibaba
Alibaba Open Source for everyone
D3
Data-Driven Documents codes.
Tencent
China tencent open source team.