Hello! This is a question about how to perform online operations after training is complete.

[HoungJee]

Congratulations! The incredible achievement you have made is truly remarkable. After running the code you provided, I have a question. Could you please guide me on how to perform online testing locally, similar to the swing demo on your homepage where punctuation tests can be conducted directly?

[qianqianwang68]

Hi, to allow for real-time visualization online without GPU like on the webpage, we saved out the dense correspondences from the trained model as png files and then visualized them in Java. If you don't want to compute the dense correspondences at once in the beginning, and don't care about being real-time too much and can test on GPUs, we also have a jupiter notebook that allows inspecting the prediction at a given clicked location. Let me know which you'd like, and I can release it after I dig it up and clean it a bit.

[HoungJee]

Hi, to allow for real-time visualization online without GPU like on the webpage, we saved out the dense correspondences from the trained model as png files and then visualized them in Java. If you don't want to compute the dense correspondences at once in the beginning, and don't care about being real-time too much and can test on GPUs, we also have a jupiter notebook that allows inspecting the prediction at a given clicked location. Let me know which you'd like, and I can release it after I dig it up and clean it a bit.

I sincerely apologize for the delayed response, and I would like to express my utmost gratitude for your answer. I believe I would prefer the second option, which allows testing on a GPU. However, the first option also seems remarkable. Please forgive my eagerness. Your work has truly fascinated me, and I am eager to learn more about this remarkable achievement. Thank you very much for your assistance.

[HoungJee]

Hi, to allow for real-time visualization online without GPU like on the webpage, we saved out the dense correspondences from the trained model as png files and then visualized them in Java. If you don't want to compute the dense correspondences at once in the beginning, and don't care about being real-time too much and can test on GPUs, we also have a jupiter notebook that allows inspecting the prediction at a given clicked location. Let me know which you'd like, and I can release it after I dig it up and clean it a bit.

Hello, after training with the processed video sequence you provided, a file named "Swing_Data_100000_0_trails.mp4" will be generated. This file displays the trajectories of tracked points. However, when I preprocess the Davis video using the method you provided and train it, the resulting trained model does not have this file. Do I need to perform visualization myself, or is there a problem with my steps?