AI Paper Atlas

Research map / Benchmarks / ScanNet v2

ScanNet v2 leaderboard

102 papers at ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR report results on ScanNet v2. Best mIoU (val): 78.52 (AnIsoNet (Ours), ICML 2026).

Numbers are as each paper reports them, read by AI from its results tables and not checked by hand. Papers can differ in setting (training data, model size, extra data); the topic column shows what each paper is about. Open the paper before you compare or quote a number.

ScanNet v2 · mIoU (val)

26 papers, each with its best reported number for its own method; higher is better.

16 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mIoU

25 papers, each with its best reported number for its own method; higher is better.

15 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mAP@0.25

20 papers, each with its best reported number for its own method; higher is better.

#mAP@0.25 ↑MethodPaperTopic
177.5H2A2
TR3D
H^2A^2: Homogeneity-Aware and Heterogeneity-Aware Feature Perception for Unified Indoor 3D Object Detection
CVPR 2026 · Table 1
3D detection and driving
275.1CAGroup3DCAGroup3D: Class-Aware Grouping for 3D Object Detection on Point Clouds
NeurIPS 2022 · Table 3D detection results
3D detection and driving
373.9OccuSegOccuSeg: Occupancy-Aware 3D Instance Segmentation
CVPR 2020 · Table, ScanNetV2 overall resul
3D generation
472.3PointGroup + Ours
PointGroup
Language-Assisted 3D Feature Learning for Semantic Scene Understanding
AAAI 2023 · 3D instance segmentation resul
3D detection and driving
570.9GroupFree3D+LG3DLabel-Guided Auxiliary Training Improves 3D Object Detector
ECCV 2022 · Table 1
Segmentation and detection
669.6RBGNet (FBS + Ray Feat 66)
VoteNet baseline
RBGNet: Ray-Based Grouping for 3D Object Detection
CVPR 2022 · Ablation table: FBS and Ray Fe
Segmentation and detection
769.3GroupFree3D*(w2×, L12, O512)+DisARM
GroupFree3D*(w2×, L12, O512)
DisARM: Displacement Aware Relation Module for 3D Detection
CVPR 2022 · Table 1
3D detection and driving
869.1GroupFree3D + HPE(MLP)
GroupFree3D
Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding
AAAI 2024 · Table 4
Segmentation and detection
969.1Group-Free (L12, O512)
PointNet++ w2x
Group-Free 3D Object Detection via Transformers
ICCV 2021 · Table 1
Segmentation and detection
1069GroupFree6,256 + X-3D
GroupFree6,256
X-3D: Explicit 3D Structure Modeling for Point Cloud Recognition
CVPR 2024 · Table 4
Point cloud classification
10 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mAP@0.25 (val)

17 papers, each with its best reported number for its own method; higher is better.

7 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mAP@0.5

17 papers, each with its best reported number for its own method; higher is better.

#mAP@0.5 ↑MethodPaperTopic
163.8H2A2
TR3D
H^2A^2: Homogeneity-Aware and Heterogeneity-Aware Feature Perception for Unified Indoor 3D Object Detection
CVPR 2026 · Table 1
3D detection and driving
263.4OccuSegOccuSeg: Occupancy-Aware 3D Instance Segmentation
CVPR 2020 · Table, ScanNetV2 overall resul
3D generation
361.3CAGroup3DCAGroup3D: Class-Aware Grouping for 3D Object Detection on Point Clouds
NeurIPS 2022 · Table 3D detection results
3D detection and driving
456.9PointGroup + Ours
PointGroup
Language-Assisted 3D Feature Learning for Semantic Scene Understanding
AAAI 2023 · 3D instance segmentation resul
3D detection and driving
554.1GroupFree3D+LG3DLabel-Guided Auxiliary Training Improves 3D Object Detector
ECCV 2022 · Table 1
Segmentation and detection
653.6GroupFree3D*(w2×, L12, O512)+DisARM
GroupFree3D*(w2×, L12, O512)
DisARM: Displacement Aware Relation Module for 3D Detection
CVPR 2022 · Table 1
3D detection and driving
753.6RBGNet (FBS + Ray Feat 66)
VoteNet baseline
RBGNet: Ray-Based Grouping for 3D Object Detection
CVPR 2022 · Ablation table: FBS and Ray Fe
Segmentation and detection
853GroupFree3D + HPE(MLP)
GroupFree3D
Improved MLP Point Cloud Processing with High-Dimensional Positional Encoding
AAAI 2024 · Table 4
Segmentation and detection
952.8Group-Free (L12, O512)
PointNet++ w2x
Group-Free 3D Object Detection via Transformers
ICCV 2021 · Table 1
Segmentation and detection
1050GroupFree (Ours)
GroupFree
A Unified Query-Based Paradigm for Point Cloud Understanding
CVPR 2022 · Indoor detection results table
Point cloud classification
7 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · AP50

16 papers, each with its best reported number for its own method; higher is better.

6 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mAP (val)

15 papers, each with its best reported number for its own method; higher is better.

5 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · mAP@0.5 (val)

15 papers, each with its best reported number for its own method; higher is better.

5 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · AP50 (val)

13 papers, each with its best reported number for its own method; higher is better.

3 more papers on this leaderboard, the baselines reported on it and a CSV download are free with a scime account.
Sign in free to see all

ScanNet v2 · AP25 (val)

10 papers, each with its best reported number for its own method; higher is better.

ScanNet v2 · mAP

10 papers, each with its best reported number for its own method; higher is better.

ScanNet v2 · AP25

10 papers, each with its best reported number for its own method; higher is better.

Not affiliated with any conference, journal or the benchmark's authors. Corrections: contact@latenticlabs.com.