metadata
license: apache-2.0
pipeline_tag: image-to-3d
tags:
- 3d-occupancy
- autonomous-driving
- computer-vision
OccAny: Generalized Unconstrained Urban 3D Occupancy
OccAny is a unified framework for generalized unconstrained urban 3D occupancy prediction. It is the first unconstrained urban 3D occupancy model capable of operating on out-of-domain uncalibrated scenes to predict and complete metric occupancy coupled with segmentation features from sequential, monocular, or surround-view images.
- Paper: OccAny: Generalized Unconstrained Urban 3D Occupancy
- Project Page: https://valeoai.github.io/OccAny
- Code: https://github.com/valeoai/OccAny
Model Variants
This repository hosts checkpoints for two model variants:
- OccAny: Based on Must3R and SAM2.
- OccAny+: Based on Depth Anything 3 and SAM3.
Sample Usage
After following the installation instructions in the GitHub repository, you can run inference using the following commands.
OccAny+ (Depth Anything 3 + SAM3)
python inference.py \
--batch_gen_view 2 \
--view_batch_size 2 \
--semantic distill@SAM3 \
--compute_segmentation_masks \
--gen \
-rot 30 \
-vpi 2 \
-fwd 5 \
--seed_translation_distance 2 \
--recon_conf_thres 2.0 \
--gen_conf_thres 6.0 \
--apply_majority_pooling \
--model occany_da3
OccAny (Must3R + SAM2)
python inference.py \
--batch_gen_view 2 \
--view_batch_size 2 \
--semantic distill@SAM2_large \
--compute_segmentation_masks \
--gen \
-rot 30 \
-vpi 2 \
-fwd 5 \
--seed_translation_distance 2 \
--recon_conf_thres 2.0 \
--gen_conf_thres 2.0 \
--apply_majority_pooling \
--model occany_must3r
Citation
If you find this work or code useful, please cite the paper:
@inproceedings{cao2026occany,
title={OccAny: Generalized Unconstrained Urban 3D Occupancy},
author={Anh-Quan Cao and Tuan-Hung Vu},
booktitle={CVPR},
year={2026}
}