unity-technologies / pysolotools Goto Github PK

In the solo2coco.py script, there is a specific method def _process_instances(...) that solely utilizes the first item within the list, and evidently, this is the limiting factor impeding the conversion process of the multi-camera SOLO Dataset.

@staticmethod
    def _process_instances(
        frame: Frame, idx, output, data_root, solo_kp_map
    ) -> Tuple[Dict, List, List, List]:
        logger.info(f"Processing Frame number: {idx}")
        image_id = idx
        sequence_num = frame.sequence
        rgb_capture = list(
            filter(lambda cap: isinstance(cap, RGBCameraCapture), frame.captures)
        )[0]

        img_record = SOLO2COCOConverter._process_rgb_image(
            image_id, rgb_capture, output, data_root, sequence_num
        )
        (
            ann_record,
            ins_ann_record,
            sem_ann_record,
        ) = SOLO2COCOConverter._process_annotations(
            image_id, rgb_capture, sequence_num, data_root, solo_kp_map
        )

        return img_record, ann_record, ins_ann_record, sem_ann_record

Are there existing methods within PySoloTools to convert a SOLO Dataset with multiple cameras into the COCO format collectively?

I tried converting a solo dataset to coco dataset. Note that the dataset had only segmentations and keypoints but no bounding boxes.

from pysolotools.converters import SOLO2COCOConverter
converter = SOLO2COCOConverter(solo)
converter.convert(output_path="path/to/output/directory", dataset_name="coco")

I am getting the following error:

` 156 if not field.init:
157 continue
--> 159 field_value = kvs[field.name]
160 field_type = types[field.name]
161 if field_value is None and not _is_optional(field_type):
KeyError: 'values'

If full error message is required, I can post it again here.

It's actually coming from solo.frames().

Hi,

I am working with a SOLO dataset where the keypoint labels are not the ones used in the COCO dataset. I would however like to convert this SOLO dataset to the COCO format. In the Keypoint Detection part of the COCO format documentation, it is explained how other keypoint names should be handled.

I would be interested in making a PR to add support for custom keypoint names. I would like to know if such a PR would be welcome, and, if yes, if there is anything I should know before starting to implement it.

(The current behavior is to simply ignore all the keypoints whose name is not in the COCO dataset, essentially failing silently.)

Hi,

Due to the settings used in my dataset generation, apparently not all images have a visible object (around 1/1000 images don't have 2D bounding boxes).
This causes the SOLO2COCOConverter to crash with a keyerror "instances" error.

Perhaps the converter can be updated to just skip images if there is no label, rather than crash? As I don't know which images don't have any bounding boxes, I now have to write a separate script which deletes these instances this before feeding the data to the converter.

I have got this error when trying to install the visualizer on Mac (the same error does not appear on windows), not sure what is causing this error, it seems to point to unity_vision.

ERROR: Could not find a version that satisfies the requirement protobuf==3.17.2 (from unity-vision) (from versions: none)
ERROR: No matching distribution found for protobuf==3.17.2

Hey,
I created a synthetic dataset using unity perception labelled only with instance segmentation,
When i was trying to convert it to COCO format i received an error of list index out of range:

Tried to dig in the code and figure it out by myself, had no success there.
Anyone as a clue why is it happening ?
When i tried to modify the code i received different errors because i couldnt recreate the data wanted by myself in this parameter.

Tnx in advance,
Mark

The Integration Test we have for Antojito/Pozole operations is using a standalone UCVD API Client to make calls.

pysolotools exists and also has nearly identical functionality. This issue is to replace the UCVD API Client used in this repo with the one from pysolotools.

pysolotools should also trigger this integration test. You can see an example of that demonstrated here.

After assigning pose labels for keypoint labeler I can see the annotation and labels in SOLO format. But after converting to COCO format any information related to pose labels and Metadata Reporter Label is missing.

Steps:

• Assign bounding box 2D, keypointLabeler and MetadataReporterLabeler on perception camera
• Assign pose config for any animation in keypointLabeler array
• create data using that animation on any humanoid model
• In SOLO format all of the above labels will be visible
• Convert this SOLO data to COCO (Issue: There will be error if you don't add box2DLabeler)
• In COCO format you can only see bbox labels and key point labels. All other labels will be missing.

Hey,

I wanted to test your tool, so I followed the README.
I must say I did pip install pysolotools without --index-url=https://artifactory.prd.it.unity3d.com/artifactory/api/pypi/pypi/simple because it wasn't working with it. But it worked anyway

When trying to run this code I got an error.

from pysolotools.converters.solo2coco import SOLO2COCOConverter
from pysolotools.consumers import Solo

solo = Solo("src_data_path")
dataset = SOLO2COCOConverter(solo)
dataset.convert(output_path="output_path")

This error kept repeating itself.

Any help on how to use your tool would be appreciated

Got another pysolotools serialization issue. Details can be found here in the logs.

KeyError: 'captures'

Generated dataset is here: gs://unity-ai-antojito-prd-dataset-files-0819/7520a09c-9447-469d-b074-ebe8705b39a1/attempt:0

Original post: https://jira.unity3d.com/browse/AIVS-969

I linked the wrong repo. boooo. the real repo is this.