search

zzangjinsun / NLSPN_ECCV20

Park et al., Non-Local Spatial Propagation Network for Depth Completion, ECCV, 2020
MIT License
321 stars 55 forks source link

Is it possible to run in real-time? #25

Closed duda1202 closed 3 years ago

duda1202 commented 3 years ago

Hello,

I would like to know if it is possible to use your code in real time. I would like to integrate it with ROS but I am not sure how it can be done based on your code. Is it possible to do it? Thank you.

zzangjinsun commented 3 years ago

Current python implementation takes 200 ms for a forward pass (+ quantitative evaluations).

I am currently working on some other works and have tried ROS node implementation on that work.

Thus I think the NLSPN ROS node can be also easily implemented, but I think it will work around 5~6 Hz with further optimization (e.g., c++ conversion) with KITTI-size images.

JUGGHM commented 3 years ago

Actually on one 2080Ti it works around 8Hz.

duda1202 commented 3 years ago

Do you know the GPU capacity while running on a 2080Ti? I was thinking of implementing this on a 2070 Super with a semantic segmentation neural network in real time. Also, thanks for your input, I will check it out how I can implement on ROS.

JUGGHM commented 3 years ago

Do you know the GPU capacity while running on a 2080Ti? I was thinking of implementing this on a 2070 Super with a semantic segmentation neural network in real time. Also, thanks for your input, I will check it out how I can implement on ROS.

About 4.2G for inference only, while the training process can be more GPU-consuming (for the original 1216x352 input it is over 10G.).