Application of neural networks in shape prediction
Is there a fixed code used in shape prediction in neural networks?
The persistent diagram (from persistent homology) can be used to get a vector that can uniquely encode shape information. check readme of this repo : https://github.com/GUDHI/TDA-tutorial/tree/master, for example you can look into this simple shape detection pipleine, https://colab.research.google.com/drive/1Dk2d7kqcVJMHWCXq1VMPS0oeYgp5-NOL?usp=sharing
Yousif Yakoub Hilal can you give more details about the problem or the domain.
Do you mean evaluating the shape of input data? Torch-summary for PyTorch or model.summary() for Tensorflow can help.
Thanks for the question. The code for shape prediction in neural networks is not fixed universally; there are common architectures and methods used nonetheless. For example, SPNet is a fully convolutional neural network for shape prediction (Xiao et al.). It is very effective in medical image segmentation. This network outputs a shape parameter vector by processing image patches and comparing those to the ground truth shapes, results optimized via layers such as convolutional layers and an Euclidean distance-based loss function. There are variations in this architecture that depend on the tasks at hand, such as object detection or more concrete scientific applications.
- Xiao, Y., Liu, D., Zhang, J., Liu, Y., & Zhan, S. (2021). SPNet: Shape Prediction Using a Fully Convolutional Neural Network. Artificial Life and Robotics, 26(1), 100–109.
- Badges
- Science topic