hkust-aerial-robotics / flow-motion-depth Goto Github PK

Dear authors, thank you very mcuh for such inspiring paper and code.
Could you please share your trained model on DeMon dataset for inference?

Thank you again!

Dear authors, I am interested in the dataset. And I would like to know if the depth maps are captured in orthographic or perspective mode?

Hi, can u pls give us any estimates of the release date?

This part of the paper is not clearly explained. Can I know how you derive them?

Thanks.

Hi, author
I am trying to figure out the theory and principle behind the code. I have some questions about the implementation of EpipolarCorrelationLayer module in correlation.py.
First, I do not understand why *10 is needed in here. Based on my understanding, *10 maybe represent the depth, but why the depth is 10?

Second, would you explain what the para_dir and perp_dir represent for. What words are they short for？

Third, from line90 to line127, I am not sure which part of your article they correspond to. It seems to be the implementation of Fig3. But I can not understand it aobut the end_point, nearest_k. Would you give a detailed explaination about that? I would appreciate you very much.

Hi, the dropbox described in README.md links to train and test sets are not working. I'm getting, Error 404 not found

Hi! I was trying to access the dataset using the provided Baidu links but was unable to download it. I would appreciate any help with regards to the same. Thank you!

Hi,

Thank you for sharing the database.
Are the depth values on the absolute scale? If yes, are they in meters or something different?

I'm planning to use your dataset for a view-synthesis-related problem I'm working on. Can you please confirm if your database will be suitable for that? To elaborate, I need to warp a frame to the view of the next frame using the relative transformation matrix, depth map, and camera intrinsic matrix.

Hi,
you gave the mean and standard deviation for the triangulation layer and flow below.