-
Notifications
You must be signed in to change notification settings - Fork 97
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
[Question] Train on 2D medical images #36
Comments
Hi @zoliseress , thanks for your interest in nnDetection. The current version does not implement a 2D pipeline because all of our data sets in the training/validation pool were 3D. Since the 2D networks produced worse results on most of our 3D data sets we did not run additional experiments in that direction. In some preliminary experiments (on pure 2D data sets), we also observed that some fixed parameters do not generalize well between 3D and 2D due to different Anchor Distributions and potentially different Network requirements (usually 2D images have a much higher resolution and thus networks with a bigger field of view could be preferable there). Best, |
Thank you for the detailed answer, Michael! I was excited about your network's performance on our data, but sadly it seems that they are incompatible. Zoli |
2D support would be of interest to me as well. I have previously experimented with RetinaNets, YOLO networks, and Faster R-CNN. Testing your design is the next step for me, to properly benchmark our designs which we have tuned ourselves. Found this rather late... I had spent quite some time getting this to install on colab with pip (working now though). Should've checked issues earlier... |
how to initialize the model? I didn't find any tips |
|
❓ Question
Hi!
Is it worth to try experimenting with the nnDetection framework on 2D medical images, or is it designed (as the description says) solely for "3D (volumetric) medical object detection"?
The text was updated successfully, but these errors were encountered: