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

AVOD pre-trained weights for mono pseudo-lidar #42

Open Pari-singh opened 4 years ago

Pari-singh commented 4 years ago

Hi @mileyan,

Thanks a lot for your code. I was trying out the monocular pipeline using your code and so wanted to know if I could use the pre-trained AVOD weights you have mentioned could be used in monocular case as well, just as mentioned in your avod_pl repository or I would need to train the AVOD model separately for the outputs from monocular depth and pseudo-lidar?

Thanks

mileyan commented 4 years ago

Hi @Pari-singh . Thanks for your interest. The pretrained model in avod_pl repo is trained on stereo data. I suggest that you train a new AVOD for your monocular depth. You can use the code in ./proprecess to project monocular depth to pseudo-lidar point cloud. Then use avod_pl to train it.