Comments (2)
Hello,
You only need to provide your own data if the current network does not work to a satisfactory level on your own data, which may deviate from the data we trained on. Have you tried running the current network on your data with the python3 segment_brain_batch.py BRAIN_FOLDER?
If you still find certain issues with a certain type of axon, you can label your own data of the axons or artifacts the network is doing poorly on. We found ImageJ's Segmentation editor (https://imagej.net/Segmentation_Editor) to be simple and easy to use.
Answer to several doubts:
Do you define a training sample as each 64x64x64 px cube with 1-2 labeled slices?
I define an ORIGINAL training example/sample as the ~128 length cubes while I call the ~64 length cubes training examples/sample.
How many training samples does the network requires if training from scratch?
If you would like to train from scratch, you will need around n = 50-100 original training samples (~128 length cubes). From these large cubes, you run python3 prepare_data.py "generate_training_set" to generate 100 * n small ~64 length cubes cropped from the larger cubes (100 training samples per an original training sample). If possible, I recommend you to not train from scratch since these original training examples would take a significant amount of effort to label.
How many training samples does the network requires for transfer learning?
Depending on how different your axons are from our training set, you will probably need around 4-10 original training examples (400-1000 training samples). You may want to include some artifact/blank chunks (examples where there are no axons) to make sure the network does not overfit to the axons.
Please let me know if you have any other questions!
from trailmap.
Hi Albert,
Thanks a lot for your answers.
No, I havent started yet, I am planning how to best process the brains. Of course, I will try first your model without further training and see how it performs.
Will most likely come back to you!
Cheers
from trailmap.
Related Issues (14)
- Install instructions HOT 2
- HDF5 file with pre-trained model weights in Github repo is corrupt HOT 7
- Blas SGEMM launch failed when predicting with GPU HOT 5
- `get_dir` gets thrown off by created `seg-` folder HOT 7
- A few questions HOT 1
- Hi, your work is good, and where can I find your training dataset? HOT 4
- Error during inference HOT 3
- Threshold and cells HOT 12
- Cuda / Cudnn
- Inconsistent environment
- note though that in contrast to your instructions I do not have h5py version 2.1 HOT 4
- VisibleDeprecationWarning for segment_brain.py HOT 1
- Model weights cannot be downloaded (`This repository is over its data quota`) HOT 1
Recommend Projects
-
React
A declarative, efficient, and flexible JavaScript library for building user interfaces.
-
Vue.js
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
-
Typescript
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
-
TensorFlow
An Open Source Machine Learning Framework for Everyone
-
Django
The Web framework for perfectionists with deadlines.
-
Laravel
A PHP framework for web artisans
-
D3
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
-
Recommend Topics
-
javascript
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
-
web
Some thing interesting about web. New door for the world.
-
server
A server is a program made to process requests and deliver data to clients.
-
Machine learning
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
-
Visualization
Some thing interesting about visualization, use data art
-
Game
Some thing interesting about game, make everyone happy.
Recommend Org
-
Facebook
We are working to build community through open source technology. NB: members must have two-factor auth.
-
Microsoft
Open source projects and samples from Microsoft.
-
Google
Google ❤️ Open Source for everyone.
-
Alibaba
Alibaba Open Source for everyone
-
D3
Data-Driven Documents codes.
-
Tencent
China tencent open source team.
from trailmap.