Scene Labelling

Deep Learning
1

Hi all , Can anybody suggest on what would be the latest with scene labelling. Do we have any examples for doing scene labelling using fast.ai ?. I have been looking at this paper “http://proceedings.mlr.press/v32/pinheiro14.pdf” . Thought to check with the forum here to get any further information on "Scene Labelling " . Thanks

2

The paper mentioned this task is using non-image segmentation method. But this was 2014. I think image segmentation method is suitable for scene labeling. If that’s the case, check out Part 2, lesson 14 where we will learn “Image Segmentation with U-Net”.

3

thanks so much. I haven’t yet reached Lesson 14. But i now know where to look at. Regards.

4

OK, good good.

5

@ cedric, do you know how to do inference using a image segmentation (U-Net) model ?. ie , how to apply predict on a new single image to get the segmented labels ?. thanks

6

I am a little busy right now. If you don’t mind, I will get back to you tomorrow.

7

Sure, thank you

8

It’s working now.

1 Like
9

Hi @cedric , i need to get the coordinates of the segmented image post applying image segmentation. Any thoughts on how this can be achieved ?

10

Sorry, I have been busy. I haven’t had time to dig into all your questions.

I am not too sure I understand your question. Anyway, here’s a quick reply:

Let’s take an example in geospatial. You can get the coordinates by translating the segmented images (map tiles) and this gives you the lat/lon bounds of that tile.

Please do not tag (@) people unless other people in the forums cannot help you, see “Etiquette for Posting to Forums”.

  1. Do not @ people if you are not referencing them for a specific reason that requires the attention of that forum member. Be specially mindful in mentioning Jeremy and Rachel.