An open source program for labeling in segmentation?

[hdnh2006]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Hello,

is there some open source program for labeling the images in the new segmenation models?

is there some documentation or common name for normalize these annotations?

Additional

No response

[glenn-jocher]

@hdnh2006 segmentation annotations are class x1 y1 x2 y2 x3 y3, ... compared to box annotations class x y w h.

You can see an example of these annotations in COCO128-seg:
https://github.com/ultralytics/yolov5/releases/download/v1.0/coco128-segments.zip

And also train on quickly with

python segment/train.py --data COCO128-seg.yaml

Note that these labels were produce by https://github.com/ultralytics/JSON2YOLO starting from COCO JSON format segmentation labels.

[barney2074]

is there some open source program for labeling the images

take a look at fiftyone/label studio/CVAT
https://voxel51.com/docs/fiftyone/integrations/labelstudio.html
https://voxel51.com/docs/fiftyone/integrations/cvat.html

Fiftone is quite sophisticated & there is a bit of a learning curve, but it is good for reviewing images & annotations, or importing/exporting data. Label Studio or CVAT by themselves are fairly straightforward.

[JoeNan1]

@glenn-jocher How to convert mask. png to data in the form of. json or. txt? At present, there are a large number of data sets whose labels are mask. png.Thanks

[hdnh2006]

Thanks @glenn-jocher , you are always awesome.

[glenn-jocher]

@JoeNan1 not sure. We don't have one ourselves for segment predictions yet, but I'm sure cv2 or numpy or maybe scikit have a binary mask to segment outline function. Other users haves asked for the same, i.e. for --save-txt labels on segmentation predictions.

[glenn-jocher]

@JoeNan1 maybe something like this:
https://scikit-image.org/docs/stable/api/skimage.segmentation.html#skimage.segmentation.find_boundaries

[glenn-jocher]

Or https://docs.opencv.org/4.1.1/d3/dc0/group__imgproc__shape.html#gadf1ad6a0b82947fa1fe3c3d497f260e0

[ryouchinsa]

You can convert the mask png file to polygons.
https://github.com/ryouchinsa/Rectlabel-support/blob/master/mask2polygons.py

You can use Labelme to label polygons, convert to the COCO format, and convert to the YOLO format.
https://github.com/wkentaro/labelme
https://github.com/ultralytics/JSON2YOLO

RectLabel is an offline image annotation tool for object detection and segmentation.
Although this is not an open source program, with RectLabel, you can read/write/export the YOLO segmentation format.

class_index x1 y1 x2 y2 x3 y3 ...
0 0.180027 0.287930 0.181324 0.280698 0.183726 0.270573 ...

[glenn-jocher]

Thank you for sharing these resources, @ryouchinsa! This will definitely be helpful for users looking to convert mask png files to polygons for segmentation labeling. Your contributions are invaluable to the YOLO community!