About the training code!!!

[huihuimaimaiti]

Thank you for your wonderful code! I am a student studying video stabilization. May I ask you for the training code? This will be very helpful to my study, and I would appreciate it if you would like to publish the training code. Thanks again!

[btxviny]

Hi I'm glad if this code was of any help. This is not the official implementation of DUT. This is the official implementation, and unfortunately it does not include the training code . As for this repo, I changed/simplified the inference code. Instead of the PWCNet for optical flow extraction (which needs custom cuda layers to be 'installed') I used RAFT which is available through torchvision and even generates better results. I also implement a different warping method inspired by PCAFlow.

[btxviny]

I have other repositories of different paper implementations including the training I used to reproduce their results. With training code:

• DMBVS
• https://github.com/btxviny/Deep-Learning-Video-Stabilization-using-Optical-Flow
• DIFRINT

Using a pretrained model: -CAIN)

And here is a repo that includes various non deep stabilization methods, using trajectory optimization and filtering. -https://github.com/btxviny/Trajectory-Optimization-and-Parametric-filtering-based-Video-Stabilization

[huihuimaimaiti]

Thank you for your patient and detailed reply. That means a lot!!!