search

mileyan / pseudo_lidar

(CVPR 2019) Pseudo-LiDAR from Visual Depth Estimation: Bridging the Gap in 3D Object Detection for Autonomous Driving
https://mileyan.github.io/pseudo_lidar/
MIT License
967 stars 215 forks source link

Have you attempted to use the depth map output by AnyNet to generate Psuedo LiDAR? #18

Open ghost opened 4 years ago

ghost commented 4 years ago

Related to #12

Just curious if your team has attempted to use the depth map output by AnyNet to generate Psuedo LiDAR?

PSMNet's performance is great but it's also quite slow on a TX2 module even at 1242x375.

I am going to trying this myself tomorrow but just curious if your team has run any experiments with this?

I was able to get AnyNet trained but without spn. Still need to write up the cpp_extension part just have not gotten to it yet.

Finally, are there any plans to release the AnyNet pre-trained model? Re: AnyNet #8

Many thanks for the great documentation and excellent work!

ghost commented 4 years ago

My apologies, I think I may have understood.

I am going to firstly run interference with AVOD model using your pre-trained weights but I am assuming it will still be quite slow on a TX2.

So I will probably have train AVOD using the output from AnyNet and see if that gives me any improvement in inference speed.

I'll update the issue here later...