yindaz / DeepCompletionRelease

Deep Depth Completion of a Single RGB-D Image
http://deepcompletion.cs.princeton.edu/
563 stars 141 forks source link

Port optimization code to python #7

Closed rohun-tripathi closed 5 years ago

rohun-tripathi commented 6 years ago

Thanks for your work! I want to port the global optimizer to python to use it for a similar task. Do you know of any python bindings for sparse solvers that I can use, or of any codebase already out there doing something similar?

yindaz commented 6 years ago

Sorry I am not familiar with Python on this. But our linear optimization is quite standard, and I believe most of the package solving linear system can do the job. Thanks.

rohun-tripathi commented 5 years ago

Thanks, I used the scipy.sparse.linalg.lsmr to solve it in python. Did you experiment with different relative weights (lamda_d and lamda_s in your paper) for the normals and the depth error components during the optimization and if so, what was your final configuration?

yindaz commented 5 years ago

Please see weights in demo: https://github.com/yindaz/DeepCompletionRelease/blob/master/matlab/demo_realsense.m

Thanks!

rohun-tripathi commented 5 years ago

Thanks for the link to the weights. I was wondering - how much does the weight for keeping the adjacent pixels at the same depth help? I tried using the system with that weight set to 0 and obtained patchy images. Have you encountered a similar output?

fuxiao567 commented 5 years ago

感谢您的工作! 我想将全局优化器移植到python,以将其用于类似的任务。您是否知道我可以使用的任何用于稀疏求解器的python绑定,或者已经存在的任何类似代码库在做类似的事情?

Thanks for your work! I want to port the global optimizer to python to use it for a similar task. Do you know of any python bindings for sparse solvers that I can use, or of any codebase already out there doing something similar?

Have you finished this work?

Chen94yue commented 2 years ago

Thanks for the link to the weights. I was wondering - how much does the weight for keeping the adjacent pixels at the same depth help? I tried using the system with that weight set to 0 and obtained patchy images. Have you encountered a similar output?

Have you finished this work?