fangchangma/sparse-to-dense

Could it be used for live depth map generation?

JeyP4 opened this issue · 3 comments

JeyP4 commented

Hi @fangchangma

Using good RGB camera and solid-state Lidar, could it be used for live applications also?

If yes, considering a good computer system(say i7-8cores | GeForce GTX 1050 Ti) what additional latency it may result?

Thanks for consideration.

fangchangma commented

Hi. I believe so. With a decent desktop GPU, the runtime can easily achieve 100+ fps, depending on which encoder/decoder you use.

Our recent ICRA'19 paper on "FastDepth: Fast Monocular Depth Estimation on Embedded Systems" addresses the latency issues specifically for embedded systems. The paper will be released on arXiv soon.

JeyP4 commented

Thanks for replying.
Are you talking about jpeg encoder/decoder.
Which one is recommended by you?

fangchangma commented

No, I was referring to the encoder/decoder in the neural network structure.