Can it stitch images in horizontal and vertical directions both. about apap-image-stitching

[EadCat]

Hi.
I think it can stitch both vertical and horizontal aligned inputs.
Please refer to "./images/demo3/results/out_00000.jpg" & "./images/demo1/results/out_00002.jpg".
However, the success of the task depends on your good placement of adjacent images and the difficulty of stitching itself.

The difficulty of stitching is related to the following factors.
#1 (comment)

[EadCat]

When real-time stitching is performed through a drone, this code is unsuitable because it uses Python's single threading.

[ChenJunzhi-buaa]

When real-time stitching is performed through a drone, this code is unsuitable because it uses Python's single threading.

Do you know any good codes or methods for real-time stitching，Thank you！

[EadCat]

Many stitching codes are not open, and there are various algorithms that assume various forms of data(Fixed view or View-free images).
All in all, the classical methods require energy optimization processes at each inference, therefore learning-based methods could be a good start if you need a real-time stitching method.
However, the learning-based methods can be limited in use due to the difficulty of building datasets.

[ChenJunzhi-buaa]

Many stitching codes are not open, and there are various algorithms that assume various forms of data(Fixed view or View-free images). All in all, the classical methods require energy optimization processes at each inference, therefore learning-based methods could be a good start if you need a real-time stitching method. However, the learning-based methods can be limited in use due to the difficulty of building datasets.

Thank for your reply! My task is to stitch DaJiang aerial video (the drone flies at an altitude of 120m) , do you know methods that can solve it well？

[EadCat]

Generally speaking, unsupervised learning-based methods will be easy to use.
Unsupervised Methods:

• UDIS: UDIS-tensorflow
  UDIS-pytorch
• HomoGAN

Semi-supervised one:

• https://github.com/megvii-research/LBHomo

If you're using fixed hardware, you can modify my code and try it, but requires a calibration process.