Execution process

[Zhongrocky]

Nice work， I have a batch of 2D blood vessel prediction results, and I want to use this post-processing method to solve the discontinuity problem. Please tell me how to use this code. I understand the need to generate training data. Then I need to train the connection model based on the training data. Finally, I need to iteratively connect the models. However, I still don’t understand the order in which this code is used to complete my post-processing tasks.

[sophie-carneiro]

Hello,

Thank you for your question! Here's how you can use the code for your task:

1. Functions for 2D Images:
   In the sources/source_2D directory, you'll find all the necessary functions for working with 2D images. The README describes what each file does.

2. Pipeline Example:
   To understand the order of operations, check out the example.py file. It demonstrates how to:

   • Create a dataset.
   • Train the neural network.
   • Apply the trained model as a post-processing step.

3. Pre-trained Model:
   If you don't have annotations or binary vascular trees to train a model, you can use the pre-trained model provided in the modeles directory.

Feel free to let me know if anything is unclear or if you encounter issues while setting it up!

[Zhongrocky]

I have a question, is the process of connecting 3D curves similar to the 2D process? I want to use this post-processing method to curve connect the results of my 3D vascular segmentation

[sophie-carneiro]

The overall pipeline (creating a dataset, training a model, and applying it iteratively) remains conceptually the same in 3D. However, the feature extraction, model design, and data representation will differ to accommodate the 3D structure. Therefore, you’ll find specific functions tailored for 3D data processing in the sources/source_3D directory and a similar example.py than in 2D.