Paintera in browser

Hi there,

Just wondering if Paintera can be used as a browser annotation tool yet? My input is expected to be 2D tiff image files and output is an integer labelled image. So far tried: Labkit (drawing option only freehand and then fill color leads to inaccurate drawings), LabelMe (no tiff file support, saves output as json n not in image formats). Paintera (is there a video tutorial for installation and usage?). Any other Python/FIji tool out there for accurate annotations? For 2D tiff FloatType -> 2D tiff IntType?

Cheers,

How about:

1 Like

I found what works for me, it is a python based tool called Labelme which requires file in png format as input but then you can draw polygons on the cell boundary and put a fill color of your choice. In the background it is making a json file of the list of co-ordinates and the fill color and with a built in command of making a dataset out of the json file creates a folder with a png file of labels. Using Fiji you cna convert such an RGB image to Luminescence image and then save it as tiff. That gives me an acurate cell annotation to train DL models on. LabelMe is a nice and easy to use tool which creates polygon based annotations. :+1:

Hi @kapoorlab, do you know this technology called HTML that allows you to add hyperlinks to your posts? :wink:

I can only suppose you mean this tool:

as opposed to LabelMe from MIT, with the source code here:

1 Like

Yup sorry, it is the first one. It is not the perfect tool I would imagine but is pretty nice as learning curve is not very steep and for small datasets is an accurate annotation tool.

1 Like

Hi @kapoorlab

I’ve been evaluating some DL systems lately and also need a good labeling tool. Thanks for this thread it was useful.

Any update as to the labeling tool you are using? It seems from your tweets that you have been doing a lot of labeling and training lately. Are you still using labelme or did you find other tools that work well?

Thanks

Brian

This other topic might be of interest (some of the 3D labeling tools discussed there work well for 2D annotations as well):

1 Like

HI @bnorthan ,

Since my data mostly comes from confocal or spinning disk these days, I use either jython scripts like these:

Roi to Label

It converts input user rois to labeled regions, or for 2D/3D data where you have some annotations this notebook using Napari can be used to correct the annotations, you can get these annotations from any computer vision or Ilastik like tools and then use this notebook to repaint:
Annotation correction

Or if you want that you paint on one slice and propagate the drawing to the next time point I use this notebook
Napari correction tool with propagation

One more thing that came out recently but I have not tried it myself is this:

Increment learning framework using Imjoy and Kaibu

1 Like