Example for dataset_meta.json "Common Semantic Segmentation (CSS)"

[andife]

Hello,

for an image segmentation task at workbench I'm trying to create a 
"Common Semantic Segmentation (CSS)" dataset. For me it's somehow unclear how to setup 
the dataset_meta.json file. Can someone provide an example dataset. 

dataset_meta.json

{

"label_map": {<map>},

"background_label":"<label>",

"segmentation_colors":[<colors>]

}

At the moment my mask (PNG file) contains three different classes, each of them is a separate channel in RGB. The background is everything which is not set in the three channels.
=> Of course I can adapt that in order to meet the requirements...

Thank you

Andreas

[JesusE_Intel]

Hi andife,

Thanks for reaching out. Take a look at a previous discussion regarding Common Semantic Segmentation (CSS) dataset. Please let me know if it answers your question.

Regards,

Jesus

[andife · Accepted solution]

The article clarifies some questions, thank you!

Is it actually also possible to specify in the format that the images should be imported as grayscale?

Additional note: my jpg file seems to be grayscale... (only 1 component)

"file pic1.jpg: JPEG image data, JFIF standard 1.01, aspect ratio, density 1x1, segment length 16, baseline, precision 8, 640x480, components 1"

[JesusE_Intel]

Hi andife,

Starting with version 2021.4, models that work with gray-scale datasets are enabled in DL Workbench. You can upload your dataset in the supported formats having grayscale images instead of 3-channel. When converting your model to IR format, you should select "Grayscale" in "Original Color Space" and the model will be converted to support input shapes that use grayscale images.

Regards,

Jesus