Very slow for 3D image

[nipunarora8]

Hi,

I installed it as a python package and the code runs super slow on 3D stack, do you have an idea how to speed it up?

Thanks

[cudmore]

Apologies for the late response.

Yes, it is slow if the start/end points are far apart.

We usually run it on start/end points about 200 pixels apart and find it acceptable.

Another issue is if the path to trace is really noisy. You could preprocess the image with a median filter or background subtraction.

Finally, we implemented a speed up by "compiling" Python with transonic but ran into issue on Windows so turned it off.

Let me know if you're on macOS or Linux and we can get you a faster version.

[nipunarora8]

Hi @cudmore
Thank you for your reply.

I somehow managed to speed it up with Numba, ~86% improvement on my Mac M3, I will make a PR (maybe in a new branch) incase you want to include the new faster version. But a fork of it is also available on my repo.

But it works flawlessly and thank you for your open sourcing it.

[cudmore]

Sounds great.

Can you link your code with numba here?

[cudmore]

Also, in the repo, you can turn on transonic for a local pip install from source.

I am curious to compare numba to transonic.

https://github.com/nipunarora8/brightest-path-lib/blob/main/brightest_path_lib/cost/__init__.py

[nipunarora8]

Hi, sorry I didn't use the transonic thing, I didn't find time to try it.
This library is a part of something I am working on. So kudos on your work.
However for more precise segmentation, I also experimented with waypoints, (additional points between start and end points).
Here is the link if it interests you: https://github.com/nipunarora8/brightest-path-lib/blob/main/brightest_path_lib/algorithm/waypointastar.py

Also, may I know what are the data samples on which this code has been tested on?
I am trying to find different photon microscopic dendrite, dendritic spine data. It would be lovely if you can share the sources.

Thanks

[jtle00]

Hi @nipunarora8,
Thank you for your work with brightest-path-lib.
I work for @cudmore in developing a GUI that uses this code. Testing your code has shown phenomenal improvements in speed, which is awesome.

I appreciate your work on waypoints as well. We have not considered that functionality before because, for our use case, it is sufficient to have our users use our GUI to choose points sequentially (one at a time within a moderate range) to create a line segment. Perhaps your waypoints can be used in the future, but nevertheless, we appreciate your contribution.

As for data samples, here are the .tif files that we have mainly been using in our development:
https://github.com/mapmanager/MapManagerCore-Data/tree/main/data/rr30a_s0u/t0
If you need anymore, please let us know.

[nipunarora8]

Hello @jtle00,
Thank you for sharing these tif files, however to make my pipeline more generic I would need more data samples. Incase you have more, I would highly appreciate if you can share them.

[jtle00]

Hi @nipunarora8, I apologize for the late response. I have shared a Google Drive of 18 data samples to your email address: nipunarora8@yahoo.com. Hopefully, that will help in developing your pipeline.

[nipunarora8]

Hello @jtle00 I highly appreciate that you sent me the data for testing my pipeline. This will help me a lot as I can make my pipeline more generic. Is it possible for you to share some metadata about the datasets, I would love to cite you guys in my report and write more about the data source, the species, anything that I can write about in my report. It's not necessary but would be a good thing to have.
Thank you once again.