ApolloScapeAuto / dataset-api

The ApolloScape Open Dataset for Autonomous Driving and its Application.
http://apolloscape.auto/
Apache License 2.0
539 stars 135 forks source link

No right images. #8

Open PeiliangLi opened 5 years ago

PeiliangLi commented 5 years ago

Hi, Thank your great effort for this good dataset. I downloaded the 3D car instance dataset, but only found the left image(camera 5). There suppose to be stereo images?

PeiliangLi commented 5 years ago

Can you tell me where to get the right image since it is a stereo dataset? @pengwangucla @dingfuzhou Thank you in advance.

pengwangucla commented 5 years ago

It is still under collection. Will get back once we obtain them. We may not have the ground truth depth every pixel, but will have stereo pairs for each image.

PeiliangLi commented 5 years ago

Thanks for your reply. I would like to evaluate my method on Apollo dataset in cvpr 2019 submission if it can be released before the ddl. Actually, I only need the rectified left-right images and extrinsic parameters.

stevenwudi commented 5 years ago

Hey @PeiliangLi , I am just curious. Do you have the performance result based on your ECCV 2018 paper (Stereo Vision-based Semantic 3D Object andEgo-motion Tracking for Autonomous Driving) : sec4 "Viewpoint Classification and 3D Box Inference" Since from my reading, you can predict the view point from a monocular RGB image based on your current existing system. Hence, I am curious about using your sec4 system, how does it perform on the current leaderboard. It is appreciated if you can report the very crude result (whether it being competitive or flawed). Thanks!

PeiliangLi commented 5 years ago

Hi, @stevenwudi , in our eccv paper, Viewpoint Classification and 3D Box Inference is a very rough result which is only based on monocular image and size prior, it's not accurate especially for far or size-variant objects. So it needs multi-steps to optimize and rectify the result to improve accuracy.

LifeBeyondExpectations commented 5 years ago

Does anyone have the stereo pair of images and LiDAR in the task of 3D Car Instance ???