Multi-Camera Bird's-Eye-View 3D Perception
Turning several camera views around a vehicle into one top-down map for 3D object detection and map segmentation.
Tier Grades
Tier A: Frontier AI labs
feasible
Shanghai AI Lab and university partners publish BEVFormer at 56.9% NDS on nuScenes.
Tier B: Indonesian AI labs
feasible
Open code and the public nuScenes benchmark make reproduction possible, though compute is not stated in the abstract.
Tier C: Institution units such as ITB
feasible
University labs can train on nuScenes with open code; compute is not stated in the abstract.
Tier D: Organisations such as SEMAI (PoT, PoC, PoV)
stretch
Multi-camera BEV perception needs a calibrated camera rig, which narrows proofs of concept to site vehicles.
Where this comes from
Tiers: a) frontier AI labs, b) Indonesian AI labs, c) institution units such as ITB, d) organisations such as SEMAI delivering proof of technology, concept and value. Each rationale cites the anchors below.
Anchors
- ComputeBEVFormer: training compute not stated in the abstract.
- DataBEVFormer: evaluated on the nuScenes test set at 56.9% NDS.
- Team size (proxy)Proxy only (author count of an exemplar paper, not a headcount): 8 authors on the BEVFormer paper (arXiv 2203.17270).
- Time horizonYears.
- Shelf-life signalsEnd-to-end driving models such as Alpamayo-R1 now plan from cameras directly, which reduces the role of a separate BEV stage.
- Pipeline roleTier A and university labs publish BEV methods; Tier B and Tier C reproduce them on local data; Tier D could use them for site vehicles such as mine trucks.
Where this comes from
Team size is a proxy (author counts of exemplar papers), not a headcount.
Who and What
Links
R&D TopicMulti-Camera Bird's-Eye-View 3D Perception
Models
BEVFormerNanjing University, driving
BEVDetPhiGent Robotics, driving
DD3DToyota Research Institute (TRI), vision
ImVoxelNetSamsung, vision+7 more
BEVFormerNanjing University, drivingBEVDetPhiGent Robotics, drivingDD3DToyota Research Institute (TRI), visionImVoxelNetSamsung, vision+7 moreUse Casesnone recorded
Who Runs Itnone recorded
Models 11
DETR3DMassachusetts Institute of Technology, Toyota Research Institute (TRI), visionproduced by
3D-Deep-BoxGeorge Mason University, Zoox, drivingproduced byBEVFormerNanjing University, Shanghai AI Lab, drivingexemplarBEVFusionMassachusetts Institute of Technology, visionproduced byDD3DToyota Research Institute (TRI), visionproduced byImVoxelNetSamsung, visionproduced by
Simple-BEVStanford University, Carnegie Mellon University, drivingproduced by
TPVFormerTsinghua University, PhiGent Robotics, visionproduced byBEVDetPhiGent Robotics, drivingexemplar
CVT (Cross-View Transformers)University of Texas at Austin, drivingproduced by
GKTHuazhong University of Science and Technology, drivingproduced byPapers
- Lift, Splat, Shoot: Encoding Images From Arbitrary Camera Rigs by Implicitly Unprojecting to 3DJonah Philion et al., ECCV 2020 (per arXiv comment), 2020 founding
- BEVFormer: Learning Bird's-Eye-View Representation from Multi-Camera Images via Spatiotemporal TransformersZhiqi Li et al., ECCV 2022 (per arXiv comment), 2022 state of the art
- BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View RepresentationZhijian Liu et al., ICRA 2023 (per arXiv comment), 2022 state of the art
- PETR: Position Embedding Transformation for Multi-View 3D Object DetectionYingfei Liu et al., ECCV 2022 (per arXiv comment), 2022 state of the art
- Delving into the Devils of Bird's-eye-view Perception: A Review, Evaluation and RecipeHongyang Li et al., arXiv preprint, 2022 survey
Where this comes from
The papers to read first: where the direction began, its current best, and a survey when one exists.
Sources
- arxiv.org/abs/2203.17270 https://arxiv.org/abs/2203.17270
To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Multi-Camera Bird's-Eye-View 3D Perception", https://aurion.semai.id/topics/multi-camera-bev-3d-perception/, records of October 2026.