# Z3D: Zero-Shot 3D Visual Grounding from Images

> A universal pipeline for localizing objects in 3D scenes from multi-view images using zero-shot methods, without geometric supervision or object priors.

## Publication record

- **Title:** Z3D: Zero-Shot 3D Visual Grounding from Images
- **Authors:** Nikita Drozdov, Andrey Lemeshko, Nikita Gavrilov, Anton Konushin, Danila Rukhovich, Maksim Kolodiazhnyi
- **Year:** 2026
- **Status:** Accepted at ACL 2026
- **Venue:** ACL 2026
- **Conference ranking:** A*-ranked conference
- **Research area:** 3D Intelligence
- **First arXiv submission:** 2026-02-03

## Abstract / summary

Z3D is a universal zero-shot 3D visual-grounding pipeline that localizes the object described by a natural-language query. It operates on multi-view images, can optionally incorporate camera poses and depth maps, and avoids task-specific geometric supervision or predefined object priors.

## Methodology

- VLM-assisted image and view selection
- Prompt-based 2D segmentation with SAM3-Agent
- Zero-shot 3D proposal generation with MaskClustering
- 2D-to-3D lifting and multi-view proposal aggregation
- Optional DUSt3R reconstruction for posed or unposed imagery

## Datasets and benchmarks

- ScanRefer
- Nr3D

## Principal results

- The paper reports state-of-the-art performance among zero-shot methods on ScanRefer and Nr3D at publication time.
- In the full ScanRefer validation setting without bounding-box or text supervision, Z3D reports overall Acc@0.25 of 54.2 and Acc@0.5 of 46.0.
- On Nr3D, Z3D reports overall top-1 accuracy of 54.8 in the paper's main zero-shot comparison.
- The same pipeline supports point-cloud, posed-RGB, and fully unposed-RGB inputs.

## Benchmark evidence

| Benchmark | Task | Input | Metric | Result | Evidence |
| --- | --- | --- | --- | --- | --- |
| ScanRefer | Zero-shot 3D visual grounding | Ground-truth point cloud; no bounding-box or text supervision | Overall Acc@0.25 / Acc@0.5 | 54.2 / 46.0 | Z3D paper, Table 1; Reported as state of the art among zero-shot approaches in the associated paper. |
| ScanRefer | Zero-shot 3D visual grounding | Posed multi-view RGB | Overall Acc@0.25 / Acc@0.5 | 42.8 / 24.8 | Z3D paper, Table 1 |
| ScanRefer | Zero-shot 3D visual grounding | Unposed multi-view RGB | Overall Acc@0.25 / Acc@0.5 | 31.2 / 12.9 | Z3D paper, Table 1 |
| Nr3D | Zero-shot 3D visual grounding | Depth-aware setting | Overall top-1 accuracy | 54.8 | Z3D paper, Table 2; The strongest zero-shot result in the paper's main Nr3D table was 54.8 versus 54.3 for SPAZER. |

Metrics are reported exactly for the stated input and evaluation setting; rows from different settings should not be treated as directly interchangeable.

## Technologies and methods

- PyTorch
- CLIP
- SAM3
- Qwen vision-language models
- DUSt3R

## Related M:3L research

- [3D Intelligence](https://m3l.am/research/3d-intelligence.md)
- [CAD Intelligence](https://m3l.am/research/cad-intelligence.md)

## External resources

- [arXiv record](https://arxiv.org/abs/2602.03361)
- [DOI](https://doi.org/10.48550/arXiv.2602.03361)
- [Code repository](https://github.com/col14m/z3d)

## Evidence boundary

Benchmark claims and numerical values above are attributed to the associated paper and its stated protocol. Potential applications are not evidence of completed deployments.

## About M:3L

M:3L is the Mathematical Modeling & Machine Learning Laboratory, a research and engineering laboratory in Yerevan, Armenia, established at the Institute of Mechanics of the National Academy of Sciences of the Republic of Armenia.

## Authoritative URLs

- [Canonical publication page](https://m3l.am/publications/z3d)
- [All M:3L publications](https://m3l.am/publications.md)
- [M:3L research](https://m3l.am/research.md)