335 - Converting COCO JSON annotations to labeled mask images

335 - Converting COCO JSON annotations to labeled masks
This video walks you through the process of converting COCO JSON annotations to labeled mask images. The four main parts of this video tutorial are:
1. Downloading data​
2. Opening the (large) JSON file in python to understand the data​
3. Visualizing a few annotations on respective images to confirm the quality of annotations​
4. Converting JSON annotations to labeled masks
Code from this video is available here: github.com/bnsreenu/python_fo...
Dataset from: github.com/sartorius-research...
Note that the dataset comes with: Creative Commons Attribution - NonCommercial 4.0 International Public License
In summary, you are good to use it for research purposes but for commercial
use you need to investigate whether trained models using this data must also comply with this license - it probably does apply to any derivative work so please be mindful.
You can directly download from the source github page. Links below.
Training json: livecell-dataset.s3.eu-central...
Validation json: livecell-dataset.s3.eu-central...
Test json: livecell-dataset.s3.eu-central...
Images: Download images.zip by following the link: livecell-dataset.s3.eu-central...
If these links do not work, follow the instructions on their github page.

[@puranjitsingh1782]

Your videos are awesome!! I have a quick question
I have a dataset in which I have images-annotations of different pixel sizes (Eg - (1800,4032), (2800, 3870) etc). In order to train the model I am dividing the images and annotations into sub-images to be used by the DL model for training. When I split the image into tiles, the annotations get distorted. Do you have any suggestions on what you think could be done?

[@fari1393]

Hi Sreeni, I have found your channel recently. Your videos are awesome. They are helping me a lot and I would like to thankyou for that. Keep up the greet job.

[@DigitalSreeni]

My pleasure 😊

[@purvi9958]

Thankyou so much!

[@nadinemuhammad6501]

Hello, thank you so much this has been very helpful🙏 I have a question, is this how the training labels for nnUnet should be like?

[@user-ei8di3om2t]

Thank you Dr. for your wonderful tutorial, i have an image that contains multiple classes, i used an apeer software to labels my image, after the labelling instead of getting a single mask with all the classes present in it, i having separate classes as masks, pls how can i convert them to a single mask

[@saqibqamar9270]

Thank you for very good tutorial. Could you make video on how to convert coco into multi class masks and train for image segmentation?

[@sulaimanmahmoud7120]

+1

[@krunalpatel7562]

++1

[@yusufozturk8509]

+111

[@TheMatrixTony]

Hey Sreeni, I've found your yt channel a few weeks ago and I really like your videos. I wanted to use your tutorial to convert the TimberSeg dataset into labeled mask images. Therefore I've tried your shared sourcecode and noticed, that objects consisting of multiple segmentations will be converted to multiple objects instead of a single object.
To fix this error I've moved the line "object_number += 1" from line 48 to line 43. In my opinion this shouldn't have any influence for the dataset that you have been using here and it works for objects consisting of multiple segmentations as well 🙂
Greetings from Germany

[@DigitalSreeni]

Thanks for sharing. :)

[@TheMatrixTony]

​@@DigitalSreeniyou're welcome! Thank you for sharing your knowledge and sourcecode with us👍

[@tektronix475]

is it COCO or LOCO json format, 4 semantic segmentation?

[@sozno4222]

Thanks!

[@DigitalSreeni]

Thank you.

[@ramu901]

Hey Sreeni - where are videos 333 and 334, what have I missed??😊

[@DigitalSreeni]

Not released yet. I inserted this video as it may provide required background for the other videos. Thanks for asking :)

[@batahena4038]

Thanks!

[@DigitalSreeni]

Thank you.