search

bostondiditeam / MV3D

Multi-View 3D Object Detection Network for Autonomous Driving
480 stars 181 forks source link

replicate results #21

Closed fferroni closed 6 years ago

fferroni commented 6 years ago

Did you try and replicate results from the original paper? Does your implementation match the paper?

myavartanoo commented 6 years ago

I have the same question?

Thanks

lancejchen commented 6 years ago

Not yet, as far as I am aware of. This repo is mainly for didi-udacity self driving car competition(Udacity provided a different dataset other than KITTI), so our goal is really to achieve better rank in the competition, not replicate results.

There are some differences between original paper and our implementation, such as you can choose resnet as base net and we haven't used exactly the same fuse net as described in the original paper.

So this repo is able to train a slightly different MV3D and run the test dataset from didi udacity competition.