RoboGeo: Benchmarking Camera-Geometry Robustness in Multi-Camera BEV Perception

WACV 2027 · Evaluations & Datasets Track. A CARLA-based benchmark that measures how robust multi-camera BEV perception models (3D detection and BEV segmentation) are to camera-geometry changes: viewpoint perturbation (VP) and cross-sensor transfer (CTS) across three ego-vehicle platforms (sedan / suv / bus).

What's inside

component	files	size	contents
metadata/carla_geobev_meta.tar.zst	1	1.5 GB	nuScenes-format DBs v1.0-carla_{sedan,suv,bus} (train, 220 scenes) + ..._eval (eval, 49 scenes), split/{train,val}.txt, detection info .pkls
platform_layouts/RGB-*.tar	18	256 GB	multi-camera RGB sweeps: 3 vehicles × 6 cameras (RGB-CAM_* = sedan, RGB-suv-CAM_*, RGB-bus-CAM_*)
viewpoint_perturbation/RGB-*.tar (+viewpoint_metadata.json)	186	~545 GB	VP evaluation captures: 6 cameras × 31 camera-pose variants (yaw/pitch/roll sweeps), 50 scenes; per-variant extrinsics in viewpoint_metadata.json

Setup

mkdir -p carla_geobev carla_VR

# metadata + images  → carla_geobev/
tar -I zstd -xf metadata/carla_geobev_meta.tar.zst -C carla_geobev
for t in platform_layouts/*.tar;  do tar -xf "$t" -C carla_geobev; done   # creates carla_geobev/sweeps/RGB-*/

# viewpoint-perturbation captures → carla_VR/
cp viewpoint_perturbation/viewpoint_metadata.json carla_VR/
for t in viewpoint_perturbation/*.tar; do tar -xf "$t" -C carla_VR; done          # creates carla_VR/sweeps/RGB-*_yaw*pitch*roll*/

The extracted carla_geobev/ is what the benchmark configs point to as the data root (nuScenes-style layout). All paths inside the DBs and .pkls are relative: the DB JSONs and carla_infos_*.pkl use sweeps/RGB-*/... (relative to carla_geobev/), while {sedan,suv,bus}_infos_*.pkl follow the mmdet3d convention data/nuscenes/sweeps/... — symlink your extracted carla_geobev/ to <repo>/data/nuscenes for those. The pkls also reference sweeps/LIDAR/*.npz entries; LiDAR files are not part of this release (camera-only evaluation never loads them — see "Not included").

Frame-rate note (important)

carla_geobev is a temporally subsampled (1/2-rate) annotation of the capture: carla_geobev frame N corresponds to carla_VR frame 2N. The viewpoint_perturbation shards here contain only the even VR frames — exactly the frames that have ground truth — so filenames in carla_VR carry frame numbers 2N.

Not included

• DPT depth maps and instance-segmentation renders (derived data; regenerable from RGB — see the benchmark code repo).
• BEV segmentation occupancy labels and raw LiDAR (available on request; num_lidar_pts / visibility attributes are already baked into the info pkls).

Benchmark

Evaluation covers 3D detection (BEVFormer, BEVDet, BEVDepth, DETR3D, PETR/CAPE, DFA3D, PD-BEV) and BEV segmentation (CVT, GaussianLSS, LaRa, LSS, PointBEV, SimpleBEV) under Normal / VP (Ext-swap, Img-swap, Cal-swap × roll/pitch/yaw) / CTS (sedan→suv, sedan→bus) conditions. Code and evaluation harness: available soon.

License & attribution

Released under CC BY-NC-SA 4.0 (non-commercial, share-alike, attribution).

This dataset was generated by replaying SimBEV (ITSC 2026) driving logs in CARLA and re-rendering all sensors. It contains no original SimBEV files; it is a SimBEV derivative released with the SimBEV authors' permission under the same CC BY-NC-SA 4.0 terms. If you use RoboGeo, please also acknowledge and cite SimBEV and CARLA:

@article{mehr2025simbev,
  title   = {SimBEV: A Synthetic Multi-Task Multi-Sensor Driving Data Generation Tool and Dataset},
  author  = {Mehr, Goodarz and Eskandarian, Azim},
  journal = {arXiv preprint arXiv:2502.01894},
  year    = {2025}
}

@inproceedings{carla,
  title     = {{CARLA}: An Open Urban Driving Simulator},
  author    = {Dosovitskiy, Alexey and Ros, German and Codevilla, Felipe and Lopez, Antonio and Koltun, Vladlen},
  booktitle = {Proceedings of the 1st Annual Conference on Robot Learning (CoRL)},
  year      = {2017}
}

Citation

@inproceedings{robogeo2027,
  title     = {RoboGeo: Benchmarking Camera-Geometry Robustness in Multi-Camera BEV Perception},
  author    = {Jeongtae Kim and Giyong Moon and Chanyoung Lee and Soonmin Hwang},
  booktitle = {Proceedings of the IEEE/CVF Winter Conference on Applications of Computer Vision (WACV)},
  year      = {2027}
}