Comments (4)
To utilize multiple GPUs during inference, generating independent segmentation for partially overlapping subvolumes is the typical thing to do. These need to be reconciled and assembled into a global segmentation. One generates ID equivalences by looking at the subvolume overlap area, computes the connected components of the resulting graph (e.g. using a union-find data structure), and writes the individual subvolumes into some volumetric storage system (e.g. using TensorStore) while relabelling the segments according to the CCs.
For training, the code is currently configured to use asynchronous SGD. One can start a process as a 'parameter server', and then some number of independent workers (one GPU each; can be on different machines) which connect to it and train together as a flock.
from ffn.
Really appreciate the reply!
I have done making some overlapping labels. I checked there were some resegmentation function in the repo. I am not sure if they are in the way you mentioned here, and I find no scripts to starting them. It would be even better if you could provide a script to run reconcile.
Thanks again for the explanation.
from ffn.
We unfortunately don't have this functionality in the main ffn repo, but I'm aware of at least one third party solution (https://github.com/Hanyu-Li/klab_utils/tree/master/klab_utils/ffn/reconciliation) which you might be able to use. IIUC, the process is to run remap.py (unique IDs per subvolume), find_graph.py (equivalences from overlapping subvolumes) and agglomerate_cv.py (update IDs according to the graph built in the previous step).
from ffn.
Really appreciate it. I will spend some time run the code and try to implement on our labels. By the way, I am wondering how you guys generate large labels. Large scale volume labels seem to be highly dependent on and limited to memory. We have got 8 Tesla-v100 GPU, but as far as I learned, it takes months to acquire the whole labels. Is this the same problem for you as well?
Kind regards.
from ffn.
Related Issues (19)
- Issue with build_coordinates.py HOT 9
- Undefined names: 'sampling' and '_required' HOT 3
- Training model does not use GPU
- Distinct samples of training data HOT 5
- syntax error in _required definition
- What about consensus and agglomeration steps? HOT 1
- Train error,maybe something wrong in bounding_box.py HOT 1
- AxisError with numpy 1.18
- TensorFlow record files are corrupted
- Parameter Optimisation Recommendations
- Combine multiple sess.run into one ?
- Share trained weights of SNEMI3D dataset?
- Qustion about blank areas in inference labels HOT 1
- Realignment and Irregular section substitution HOT 4
- TypeError: string indices must be integers HOT 1
- Evaluation the segmentation HOT 1
- build_coordinates generating TFRecords so slowly
- ffn_inference_colab_demo issue 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 ffn.