BhopalMLS: A LiDAR Dataset for Semantic Scene Understanding in Heterogeneous Urban Environments
Organizations: GeoAI4Cities Lab, Department of Data Science and Engineering, IISER Bhopal, India · Department of Civil Engineering, IIT Kanpur, India
Abstract
We introduce BhopalMLS, a large-scale Mobile Laser Scanning (MLS) dataset from a dense, heterogeneous urban environment in Bhopal, India, with mixed traffic, informal structures, and narrow unplanned streets. It spans approximately 10 km of road corridors with over 1.2 billion points and 19,518 annotated instances across 62 fine-grained static categories, grouped into 12-class and 21-class benchmark taxonomies; 6 optional moving-class subclasses support dynamic-object analysis. Existing 3D datasets, collected in well-planned cities, are structurally more uniform than such unplanned settings, and lack their subtype diversity. On semantic and instance segmentation at both granularities, fine-tuned foundation models surpass supervised baselines: Utonia(L) reaches 71.28% mIoU (vs. PTv3's 68.04% from scratch) and 44.07% mAP50 on the 12-class benchmark. Cross-region experiments against four established datasets (Ke-MLS, Toronto-3D, Paris-Lille-3D, SemanticKITTI) reveal a consistent asymmetric transferability: models trained on BhopalMLS transfer better than the reverse on every pair, for both Utonia(L) and Concerto(L) (mean 21.5, up to 36 mIoU points), holding even when the target is denser. Factor-controlled retraining at a fixed label budget, repeated on both models, attributes this to the coverage of BhopalMLS's subtype inventory rather than to dataset size or corridor geometry: collapsing each class to a single subtype costs 13.0 and 13.3 mIoU points for Utonia(L) and Concerto(L). BhopalMLS therefore offers a testbed for 3D perception beyond well-planned cities. Data and an interactive viewer are at https://geoai4cities.github.io/bhopalmls.
Figures & tables
| Name | Year | Points | Classes | RGB | Sensor | Inten- sity | Inst- ance | Country |
|---|---|---|---|---|---|---|---|---|
| Oakland [ 17 ] | 2009 | 1.6M | 5 (44) | MLS | USA | |||
| Paris-rue-Madame [ 21 ] | 2014 | 20M | 17 | MLS | France | |||
| TUM-MLS [ 36 ] | 2016 | 40M | 8 | MLS | Germany | |||
| Semantic3D [ 6 ] | 2017 | 4,000M | 8 | TLS | Europe | |||
| Paris-Lille-3D [ 20 ] | 2018 | 143M | 9 (50) | MLS | France | |||
| SemanticKITTI [ 3 ] | 2019 | 4,549M | 19 (28) | MLS | Germany |
| Major Class | Inst. | Pts (M) | Pts (%) |
|---|---|---|---|
| Building | 1,326 | 813.0 | 66.0 |
| Road | 289 | 157.0 | 12.8 |
| Vehicle | 9,888 | 77.2 | 6.3 |
| Wire | 522 | 39.0 | 3.2 |
| Vegetation (Veg.) | 374 | 33.6 | 2.7 |
| Road Side Object (RSO) | 2,294 | 30.4 | 2.5 |
| Indian MLS | Western | ||||
| Metric | BM | KM | PL | TR | SK |
| (Ours) | |||||
| Sensor | handheld | vehicle | vehicle | vehicle | spinning |
| ZEB | Riegl VQ-450 | Velo-32 | MLS | 64-ch | |
| Region | India | India | France | Canada | Germany |
| Density | |||||
| Model | OA (%) | mAcc (%) | mIoU (%) | Board | Building | Fencing | Ground | Human&Animal | Pole&Tower | Road | Road Side Obj. | Vegetation | Vehicle | Wire | Wire-Structure Conn. |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Supervised Learning – 4ch (XYZ + intensity) | |||||||||||||||
| PTv3 [ 30 ] | 92.60 | 78.75 | 67.88 | 57.05 | 92.75 | 13.87 | 22.24 | 66.09 | 69.41 | 87.39 | 51.03 | 97.11 | 91.19 | 87.74 | 78.75 |
| LitePT-L [ 33 ] | 91.13 | 80.45 | 67.53 | 57.19 | 91.47 | 20.24 | 22.10 | 63.88 | 66.69 | 85.07 | 48.76 | 97.18 | 88.23 | 86.06 | 83.47 |
| KpConvX-L [ 26 ] | 92.00 | 80.10 | 66.78 | 55.82 | 92.31 | 20.74 | 24.28 | 60.30 | 62.87 | 86.20 | 49.12 | 97.24 | 89.40 | 86.25 | 76.80 |
| Supervised Learning – 9ch (XYZ + Partial RGB + normals) | |||||||||||||||
| PTv3 | 92.17 | 79.57 | 68.04 | 59.37 | 92.42 | 18.18 | 22.69 | 62.22 | 68.23 | 86.65 | 49.37 | 97.25 | 90.78 | 86.61 | 82.71 |
| Source | Utonia(L) | Concerto(L) | ||||
| Category | Target | mAcc | mIoU | mAcc | mIoU | |
| Within- domain | BM BM | 12 | 83.89 | 71.28 | 80.44 | 71.06 |
| KM KM | 12 | 77.05 | 66.66 | 74.79 | 67.58 | |
| TR TR | 7 | 93.25 | 88.38 | 93.25 | 86.96 | |
| PL PL | 11 | 92.11 | 84.77 | 92.03 | 83.43 | |
| SK SK | 9 | 84.86 | 77.14 | 83.67 | 75.09 | |
Appendix figures & tables26 assets
Supplementary material from the paper’s appendix.
Appendix
| Major Classes (Part 1) | Major Classes (Part 2) | ||||||||
| Major Class | Subclass | Inst. | Pts(M) | Pts(%) | Major Class | Subclass | Inst. | Pts(M) | Pts(%) |
| Building | Total | 1,326 | 813.0 | 66.0 | Human & Animal | Total | 8,864 | 22.5 | 1.8 |
| —– Commercial | 724 | 628.8 | 51.1 | —– Moving Humans † | 5,386 | 12.6 | 1.0 | ||
| —– Residential | 106 | 84.0 | 6.8 | —– Human & Animal | 2,005 | 4.9 | 0.4 | ||
| —– Building (gen.) | 439 | 52.6 | 4.3 | —– Male | 996 | 3.2 | 0.3 | ||
| —– Special | 56 | 47.2 | 3.8 | —– Female | 294 | 1.1 | 0.1 | ||
| Path | Split | Len. (m) | Scenes | Pts (M) | Turn ( ∘ /100m) | Width (m) | Bld. % | Road % | Veg. % | Veh. % |
|---|---|---|---|---|---|---|---|---|---|---|
| 1 | train | 1,663 | 418 | 137.2 | 56.1 | 3.39 | 57.5 | 16.6 | 5.1 | 5.9 |
| 2 | train | 1,846 | 270 | 147.4 | 82.5 | 3.47 | 68.8 | 13.9 | 1.0 | 6.9 |
| 3 | train | 1,110 | 331 | 98.9 | 46.2 | 3.20 | 61.8 | 14.9 | 5.2 | 7.4 |
| 4 | train | 1,845 | 334 | 156.8 | 82.3 | 3.32 | 60.6 | 15.2 | 3.9 | 6.2 |
| 5 | val | 1,452 | 380 | 143.2 | 74.4 | 2.69 | 72.9 | 11.2 | 1.0 | 3.2 |
| 6 | val | 657 | 188 | 83.4 | 77.5 | 2.59 | 74.3 | 10.6 | 0.1 | 4.2 |
| Class | Share (%) | Agreement (%) | Model IoU |
|---|---|---|---|
| Wire-Structure Conn. | 2.8 | 99.8 | 83.0 |
| Building | 64.0 | 96.4 | 93.4 |
| Vegetation | 1.3 | 94.6 | 97.4 |
| Road | 21.3 | 94.1 | 85.7 |
| Wire | 2.1 | 92.2 | 87.5 |
| Road Side Object | 2.1 | 88.1 | 51.5 |
| val | test | |
|---|---|---|
| within 5 m of train | 0 6.7% | 35.0% |
| within 10 m of train | 0 7.6% | 39.7% |
| within 25 m of train | 10.4% | 51.8% |
| median distance to train | 89.7 m | 22.7 m |
| observed-area overlap | 15.7% | 49.8% |
| Model | OA (%) | mAcc (%) | mIoU (%) | Ground | Stairs | Vegetation | Building (other) | Commercial | Road | Wire | Pole (other) | Electric Pole | Street Light | 2-Wheeler | 3-Wheeler | 4-Wheeler | Hand Cart | Board | Fencing | Human | Road Side Furniture | Shop Shade | Mannequin | Road Side Obj (other) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Supervised Learning – 4ch (XYZ + intensity) | ||||||||||||||||||||||||
| PTv3 [ 30 ] | 86.47 | 71.13 | 58.35 | 16.80 | 14.25 | 97.04 | 30.44 | 84.02 | 87.08 | 86.61 | 70.47 | 66.03 | 58.03 | 95.09 | 86.28 | 83.00 | 62.66 | 60.82 | 21.06 | 65.08 | 38.43 | 58.90 | 28.56 | 14.78 |
| LitePT-L [ 33 ] | 84.73 | 71.70 | 55.09 | 16.67 | 13.51 | 96.52 | 29.57 | 82.25 | 86.53 | 85.08 | 65.95 | 64.61 | 52.38 | 93.57 | 85.79 | 73.33 | 47.63 | 51.39 | 22.48 | 63.32 | 34.26 | 57.10 | 25.18 | 9.83 |
| KpConvX-L [ 26 ] | 84.09 | 71.23 | 54.47 | 15.96 | 13.59 | 97.03 | 32.13 | 81.21 | 85.34 | 85.66 | 68.74 | 62.49 | 47.15 | 92.24 | 74.84 | 77.13 | 51.69 | 55.42 | 20.36 | 60.75 | 36.74 | 61.08 | 19.82 | 4.47 |
| Supervised Learning – 9ch (XYZ + Partial RGB + normals) | ||||||||||||||||||||||||
| PTv3 | 86.61 | 70.70 | 58.51 | 18.40 | 13.51 | 96.79 | 28.15 | 84.19 | 87.06 | 84.76 | 71.14 | 66.89 | 56.15 | 94.16 | 88.92 | 84.17 | 64.74 | 57.27 | 22.24 | 65.89 | 38.83 | 59.68 | 31.55 | 14.20 |
| Model | Setting | OA (%) | mAcc (%) | mIoU (%) | Board | Building | Fencing | Ground | Human&Animal | Pole&Tower | Road | Road Side Obj. | Vegetation | Vehicle | Wire | Wire-Structure Conn. |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Concerto(L) | (C0) full recipe | 93.42 | 80.44 | 71.06 | 67.72 | 93.71 | 20.14 | 25.41 | 69.88 | 72.48 | 87.11 | 53.07 | 97.67 | 92.34 | 87.91 | 85.26 |
| (C1) no class-weights | 93.46 | 77.84 | 69.44 ( 1.62) | 62.52 | 93.63 | 8.20 | 23.52 | 71.07 | 71.98 | 87.29 | 50.74 | 97.64 | 91.80 | 88.62 | 86.29 | |
| (C2) no Lovász | 93.03 | 81.22 | 70.67 ( 0.39) | 66.45 | 93.26 | 22.50 | 25.04 | 69.08 | 71.93 | 86.62 | 50.74 | 97.69 | 92.69 | 87.39 | 84.64 | |
| (C3) CE only (C1 C2) | 93.52 | 78.16 | 70.21 ( 0.85) | 68.09 | 93.79 | 14.63 | 24.52 | 68.66 | 71.76 | 87.42 | 50.17 | 97.62 | 92.20 | 88.81 | 84.80 | |
| (C4) no Building downsampling | 93.48 | 79.64 | 70.44 ( 0.62) | 64.74 | 93.81 | 21.27 | 23.51 | 69.63 | 72.85 | 86.48 | 48.83 | 97.76 | 92.84 | 88.11 | 85.44 | |
| Utonia(L) | (C0) full recipe | 92.69 | 83.89 | 71.28 | 67.36 | 93.38 | 38.79 | 26.27 | 64.24 | 68.61 | 85.66 | 51.51 | 97.41 | 91.54 | 87.53 | 83.03 |
| Model | mAP25 (%) | mAP50 (%) | mAP (%) | Building | Pole & Tower | Wire-Structure Conn. | Vehicle | Board | Fencing | Human & Animal | Road Side Object |
|---|---|---|---|---|---|---|---|---|---|---|---|
| Supervised Learning – 9ch (XYZ + Partial RGB + normals) | |||||||||||
| PTv3 [ 30 ] | 42.07 | 35.13 | 25.78 | 23.71 | 28.57 | 46.14 | 70.62 | 41.98 | 0.13 | 54.04 | 15.87 |
| LitePT-L [ 33 ] | 43.73 | 36.76 | 26.90 | 34.03 | 39.15 | 30.54 | 69.43 | 45.87 | 3.86 | 54.84 | 16.33 |
| Linear Probe (9ch: XYZ + Partial RGB + normals) | |||||||||||
| Sonata(L) [ 31 ] | 23.28 | 14.66 | 7.72 | 20.41 | 2.33 | 13.04 | 49.87 | 6.72 | 2.96 | 19.27 | 2.63 |
| Concerto(O) | 11.86 | 4.07 | 1.66 | 4.41 | 1.83 | 0.00 | 2.35 | 8.39 | 0.01 | 14.86 | 0.75 |
| Model | mAP25 (%) | mAP50 (%) | mAP (%) | 2-Wheeler | 3-Wheeler | 4-Wheeler | Hand Cart | Human | Commercial Building | Building (other) | Electric Pole | Street Light | Pole (other) | Board | Fencing | Road Side Furniture | Shop Shade | Mannequin | Stairs | Road Side Object (other) |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Supervised Learning – 9ch (XYZ + Partial RGB + normals) | ||||||||||||||||||||
| PTv3 [ 30 ] | 39.06 | 33.07 | 24.31 | 68.43 | 85.20 | 59.22 | 55.28 | 52.55 | 34.00 | 6.69 | 50.79 | 25.78 | 18.92 | 43.58 | 1.57 | 19.22 | 24.43 | 11.93 | 2.08 | 2.43 |
| LitePT-L [ 33 ] | 33.26 | 25.81 | 18.58 | 70.65 | 73.15 | 42.54 | 50.67 | 55.02 | 0.00 | 0.15 | 20.94 | 39.06 | 2.32 | 28.94 | 1.33 | 15.00 | 24.92 | 7.25 | 1.72 | 5.17 |
| Linear Probe (9ch: XYZ + Partial RGB + normals) | ||||||||||||||||||||
| Sonata(L) [ 31 ] | 24.37 | 16.07 | 9.46 | 52.38 | 52.75 | 46.74 | 32.08 | 19.80 | 40.76 | 1.41 | 3.88 | 2.40 | 3.35 | 2.83 | 2.87 | 2.52 | 7.99 | 0.40 | 1.09 | 0.00 |
| Concerto(O) | 11.44 | 4.44 | 1.73 | 1.03 | 13.00 | 7.57 | 11.12 | 1.08 | 15.87 | 0.75 | 2.23 | 4.70 | 0.94 | 11.22 | 0.30 | 0.07 | 5.56 | 0.02 | 0.00 | 0.00 |
| Model (train test) | OA (%) | mAcc (%) | mIoU (%) | Board | Building | Fencing | Ground | Human&Animal | Pole&Tower | Road | Road Side Obj. | Vegetation | Vehicle | Wire | Wire-Structure Conn. |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Within-domain (source target) | |||||||||||||||
| Concerto(L) (BM BM) | 93.42 | 80.44 | 71.06 | 67.72 | 93.71 | 20.14 | 25.41 | 69.88 | 72.48 | 87.11 | 53.07 | 97.67 | 92.34 | 87.91 | 85.26 |
| Utonia(L) (BM BM) | 92.69 | 83.89 | 71.28 | 67.36 | 93.38 | 38.79 | 26.27 | 64.24 | 68.61 | 85.66 | 51.51 | 97.41 | 91.54 | 87.53 | 83.03 |
| Concerto(L) (KM KM) | 92.40 | 74.79 | 67.58 | 13.83 | 83.47 | 83.66 | 68.43 | 76.87 | 76.15 | 86.04 | 0.00 | 96.59 | 90.15 | 86.69 | 49.05 |
| Utonia(L) (KM KM) | 92.82 | 77.05 | 66.66 | 13.54 | 84.27 | 86.85 | 69.26 | 83.28 | 72.43 | 87.33 | 0.00 | 96.48 | 94.47 | 87.55 | 24.43 |
| Concerto(L) (TR TR) | 99.03 | 93.25 | 86.96 | 0.00* | 93.71 | 50.18 | 0.00* | 0.00* | 85.41 | 99.41 | 0.00* | 97.45 | 95.61 | 86.94 | 0.00* |
| Arm | Removes | Scenes | Lab. pts | In-dom. |
|---|---|---|---|---|
| full | nothing | 1,353 | 540.4M | 71.28 |
| (B1) size | 75% of scenes, at random | 360 | 136.1M | 67.99 |
| (B2) variety | non-dominant subclass labels | 431 | 135.2M | 61.18 |
| (B3) geometry | all but regular corridors | 295 | 135.5M | 68.61 |
| (B4) subtype | as B2, Vehicle keeps Car | 447 | 135.2M | 56.34 |
| In-dom. | KM | TR | PL | SK | |
| full model, BM X | 71.28 | 56.78 | 77.27 | 55.68 | 40.05 |
| reverse, X BM | – | 33.96 | 51.65 | 27.06 | 35.54 |
| asymmetry gap | – | 22.82 | 25.62 | 28.62 | 0 4.50 |
| (B1) size | 67.99 | 52.56 | 76.17 | 48.87 | 47.94 |
| (B2) variety | 61.18 | 42.22 | 61.63 | 34.90 | 26.28 |
| (B3) geometry | 68.61 | 49.30 | 78.70 | 52.56 | 46.01 |
| In-dom. | KM | TR | PL | SK | |
| full model, BM X | 71.06 | 49.23 | 75.20 | 48.84 | 32.07 |
| reverse, X BM | – | 32.70 | 39.11 | 31.07 | 12.29 |
| asymmetry gap | – | 16.52 | 36.09 | 17.76 | 19.78 |
| (B1) size | 68.52 | 50.50 | 72.34 | 49.75 | 30.45 |
| (B2) variety | 59.38 | 39.74 | 60.10 | 32.71 | 16.82 |
| (B3) geometry | 68.12 | 50.44 | 72.63 | 49.48 | 33.07 |
| Arch. | Method | Ground | Vegetation | Building | Road | Wire | Pole&Tower | Vehicle | Board | Fencing | Human&Animal | Road-Side-Obj | mIoU 11 | mIoU 7 |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Concerto | vanilla | 12.39 | 60.33 | 95.82 | 75.02 | 50.72 | 25.23 | 96.39 | 20.30 | 21.37 | 19.86 | 59.76 | 48.84 | 60.70 |
| retrain | 25.92 | 73.61 | 91.29 | 64.36 | 17.74 | 29.17 | 90.04 | 13.31 | 15.98 | 34.63 | 30.98 | 44.28 | 54.60 | |
| + CORAL | 25.12 | 70.57 | 87.39 | 64.99 | 6.90 | 26.96 | 87.86 | 12.96 | 11.04 | 24.63 | 35.14 | 41.23 | 50.81 | |
| Utonia | vanilla | 28.07 | 65.48 | 98.04 | 78.26 | 48.53 | 41.00 | 97.76 | 24.36 | 30.82 | 52.87 | 47.28 | 55.68 | 65.70 |
| retrain | 38.27 | 75.08 | 94.37 | 68.74 | 24.39 | 36.46 | 90.48 | 28.28 | 41.79 | 59.43 | 26.31 | 53.06 | 61.62 | |
| + CORAL | 31.23 | 74.32 | 90.52 | 66.69 | 14.60 | 33.55 | 88.22 | 25.69 | 39.87 | 32.51 | 21.74 | 47.18 | 58.25 |
3D Scene Understanding
3D scene understanding builds geometric and semantic representations of environments for spatial reasoning, perception, and interaction.
See all 307 papers