-
Notifications
You must be signed in to change notification settings - Fork 25
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
output very weird #82
Comments
Modern arts... puh good question... When zooming in a little, what are these structures? Are these boxes around each pixel? Did you train the model yourself? what was the input? |
Thank you for your reply and sorry for the late reply! This is an optic nerve section with a nuclear marker. I checked my output and there weren't boxes around each pixel. I trained the model based on the pre-trained detection model provided on the website. The crosses are targeting the background of the tissue instead of the actual signal, and I think I might labelled my images incorrectly. Do you have more detailed instructions on how to annotate images on imageJ and dos and don'ts? That will be extremely helpful! |
@ThorstenFalk Sorry I forget to mention you in my reply! |
Ah it's a detection output, so it is probably crosses on every other background pixel. That happens if the network for whatever reason predicted a checkerboard pattern. Nevertheless, the output you get is very strange and looks more like a problem with input data. What's the element size of your images? Is it properly set in ImageJ? Did you try segmentation instead of detection? What's the output? You can also enable intermediate network outputs (i.e. heatmaps) for further debugging. If you see massive checkerboarding there, it indicates that your model is actually untrained or really badly trained. Actually there is no specifically pre-trained detection model. However, when you use a segmentation model in a detection job it should report the centers of gravity of the segmented regions, which is pretty much what you want. However, without finetuning, results will probably be poor. |
Hi Thorsten! Thank you so much for your suggestion! I will try segmentation instead. I will let you know once I tried. |
Hi everyone,
![image](https://user-images.githubusercontent.com/45889981/138903719-6a7c88ed-2a55-476f-9c33-cc1fef97c347.png)
I set up the u-net on aws ec2 and its running now. However, my result looked very off:
and on the result the output is not the name that I annotated my images. Does any of you guys know what happened and how to resolve?
Thank you kindly!
The text was updated successfully, but these errors were encountered: