1
$\begingroup$

I've got some images of cell organelles and I really want to avoid labeling them by hand. The images all look something like the image below. Is there already an existing model specifially for cells? I tried it with existing models like mask_rcnn_coco.h5 or deeplabv3_xception65_ade20k.h5 but as they aren't built for cells the results aren't good.

My code is pretty basic right now:

from pixellib.instance import instance_segmentation

from PIL import Image

    def segmentImage():
        segment_image = instance_segmentation()
    
        segment_image.load_model("mask_rcnn_coco.h5")
        segment_image.segmentImage("cell.jpg", output_image_name="cell_seg.jpg", show_bboxes=True)

Cell example image:

cell organelle example image

Example segmentation result:

enter image description here

$\endgroup$
2
$\begingroup$

I am the creator of PixelLib. PixelLib supports instance segmentation of 80 classes of objects using pre-trained coco model. Cell organelles is not part of the classes supported. The only solution is to train your own custom model. PixelLib supports custom training of a model using your own labelled dataset. Read this article I published on how to train a custom model with PixelLib.

$\endgroup$
2
  • $\begingroup$ Wow, what an honor that you personally respond. Thank you very much. I actually already found this article and I wanted to avoid labelling the cells by hand. But I guess it pays off in the long run hopefully. I see a lot of potential in your work for Bioinformatics in general and especially for the project i'm working on now. If done right I'm sure PixelLib can help save us a lot of work :) $\endgroup$ – Cowboy_Patrick Dec 11 '20 at 9:53
  • $\begingroup$ Thank you...... $\endgroup$ – Ayoola Olafenwa Dec 12 '20 at 10:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.