CVPR Daily - Thursday

The team have also been organizing a challenge, the Dense Depth for Autonomous Driving (DDAD) challenge , which deals with depth estimation from monocular images, within the context of monocular 3D perception. The DDAD dataset is a new dataset that was collected by TRI vehicles and has been released publicly. They hope it is going to help push the current state of the art in monocular depth estimation forward and complement some of the existing datasets with longer-range LiDAR and denser points, which will allow people to estimate in a more fine-grained fashion the performance of monocular depth estimation. The winners will be announced during the workshop and will have an opportunity to claim their prize and present their methods. The workshop was originally intended to run last year but was postponed. “ We had planned to organize this workshop in person last year, but we delayed it due to Covid, ” Igor reveals. “ We had a lot of the groundwork already done. We were really hoping this year it would be in person, but we’re all still virtual. We hope it will be different next time. ” We wish the team every success and look forward to hearing that they will be back next year to organize their second workshop in person, with even more interesting thoughts and learnings for the community. In the meantime, you are all invited to join the Frontiers of Monocular 3D Perception workshop at virtual CVPR on Friday morning. 16 DAILY CVPR Thursday Workshop PackNet architecture (open-sourced), proposed at CVPR2020, applied to the DDAD dataset Results of monocular 3D object detection network (it regresses bounding boxes directly from an RGB image)