Research map / Detection and segmentation / 3D detection and driving
DEVIANT: Depth EquiVarIAnt NeTwork for Monocular 3D Object Detection
Abhinav Kumar, Garrick Brazil, Enrique Corona, Armin Parchami, Xiaoming Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace vanilla convolutions with scale-equivariant SES convolutions | GUP Net · KITTI Val cars | AP3D|R40 Moderate (%): 15.48 -> 16.54 | Improved |
| add scale equivariance on top of scale augmentation (ablation) | GUP Net · KITTI Val cars | AP3D|R40 Easy (%): 21.10 -> 24.63 | Improved |
| use SES convolution instead of log-polar convolution (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 6.77 -> 16.54 | Improved |
| use SES convolution instead of dilated convolution (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 15.49 -> 16.54 | Improved |
| use SES convolution instead of DISCO (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 13.84 -> 16.54 | Improved |
| use floating downscaling alpha 10% instead of 5% (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 16.51 -> 16.54 | Improved |
| use floating downscaling alpha 10% instead of 20% (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 15.85 -> 16.54 | Improved |
| make batch norm and pooling scale-equivariant too (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 16.20 -> 16.54 | Improved |
| use three Hermite scales instead of one (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Hard (%): 13.63 -> 14.52 | Improved |
| use three Hermite scales instead of two (ablation) | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 16.48 -> 16.54 | Improved |
| replace scale-equivariant backbone with DETR3D transformer | DEVIANT · KITTI Val cars | AP3D|R40 Moderate (%): 1.26 -> 16.54 | Improved |
| use DEVIANT backbone instead of vanilla | GUP Net · KITTI Val to nuScenes frontal Val | depth MAE (all): 1.45 -> 1.26 | Improved |
Later papers that compared with this method
7 later accepted papers report DEVIANT as a baseline in their results tables; 6 of them report a better number on most of the results they share with it.
- Unleashing the Power of Chain-of-Prediction for Monocular 3D Object Detection (CVPR 2026 · better on 1 of 1 shared result)
- RARE: Learn to RAnk and REtrieve for Monocular 3D Object Detection (CVPR 2026 · better on 1 of 1 shared result)
- MonoSOWA: Scalable Monocular 3D Object Detector Without Human Annotations (ICCV 2025 · better on 1 of 1 shared result)
- MonoMAE: Enhancing Monocular 3D Detection through Depth-Aware Masked Autoencoders (NeurIPS 2024 · better on 1 of 2 shared results)
- MonoUNI: A Unified Vehicle and Infrastructure-side Monocular 3D Object Detection Network with Sufficient Depth Clues (NeurIPS 2023 · better on 1 of 1 shared result)
- MonoNeRD: NeRF-like Representations for Monocular 3D Object Detection (ICCV 2023 · better on 1 of 1 shared result)
- Monocular 3D Object Detection with Bounding Box Denoising in 3D by Perceiver (ICCV 2023 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Categorical Depth Distribution Network for Monocular 3D Object Detection (CVPR 2021 · similarity 0.87)
- Transformation-Equivariant 3D Object Detection for Autonomous Driving (AAAI 2023 · similarity 0.86)
- Rotationally Equivariant 3D Object Detection (CVPR 2022 · similarity 0.86)
- Triangulation Learning Network: From Monocular to Stereo 3D Object Detection (CVPR 2019 · similarity 0.86)
- MonoCD: Monocular 3D Object Detection with Complementary Depths (CVPR 2024 · similarity 0.86)
- MonoDGP: Monocular 3D Object Detection with Decoupled-Query and Geometry-Error Priors (CVPR 2025 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
