Awesome

Registration_CNN_NTG

Paper

• "A Multispectral Image Registration Method Based on Unsupervised Learning"(基于无监督学习的多光谱图像配准)

Poster

For the speed of the web loading, the image of the poster has been put in the link(poster-compressed-public.jpg)

Train

The average training progress costs about 6 hours.

• Train.py: The geometric-model we used is the affine model and the training data is generatered online through random affine params.
• model/cnn_registration_model.py: the main model of our model, which contains feature extraction, feature matching, feature regression.
• ntg_pytroch/register_loss.py: This file contains our unsupervised loss function, which is first proposed by this paper.

Test

• multispectral_pytorch_batch.py: We use two-stage registeration progress to achieve the sub-pixel level accuracy. Firstly, the deep model is used to estimate the rough affine params. Then we will use the traditional ntg method to optimize the rough params.

Visualization

For the purpose of visualization, we add the pyqt client to use our method quickly.

FAQ

If you have other questions, welcone to submit issues.