LIV-GaussMap
LIV-GaussMap 以硬體同步的 LiDAR-慣性系統及尺寸自適應體素取得位姿與平面結構,將體素平面的共變異轉為高斯初始形狀,再用影像光度梯度精修球諧顏色與結構。作者在 FusionPortable 以 Chamfer、EMD 與 F-score 評估結構,並報告光度最佳化會使結構品質略為下降,顯示渲染與幾何之間的取捨。
本頁內容
Initializes surface Gaussians from LiDAR-inertial adaptive-voxel planes and refines them photometrically; photometric refinement slightly degrades structure metrics.
技術屬性
欄位內容為文獻擷取紀錄的原文用語(英文),以原文為據;「未查證」表示本研究尚未讀到該資訊,不代表該方法不具備此能力。
| 感測輸入 | 3D LiDAR (Livox Avia; Ouster OS1-128; solid-state RealSense L515)、IMU、monocular camera (global shutter; rolling shutter on the L515) |
|---|---|
| 原文測試平台 | 未記錄 |
| 狀態估計 | LiDAR-inertial odometry with size-adaptive voxel map provides poses; Gaussians refined by photometric gradients |
| 資料關聯 | LiDAR plane covariances for Gaussian initialization; photometric gradients for refinement |
| 時間表示 | discrete poses |
| 去畸變 | 不適用 |
| 迴圈閉合 | none reported |
| 全域最佳化 | none reported |
| 地圖表示 | surface Gaussians initialized from size-adaptive voxel plane covariances; spherical-harmonic colour |
| 先驗資訊 | none |
| 可輸出幾何 | Gaussian map with renderings; structure evaluated against ground-truth point clouds (CD, EMD, F-score) |
| 計算需求 | Intel i9-12900K + RTX 4090 (Sec. IV) |
使用設備
原文使用的感測器、運算硬體與載具(equipment)。型號保留原文寫法,連結到設備頁中同一型號的歸併名稱;角色依原文用途分為方法輸入、資料集感測器、執行運算平台、參考或真值量測(reference or ground truth)與比較對象設備。
| 類別 | 型號(原文寫法) | 角色 | 資料集 | 原文規格 | 出處 |
|---|---|---|---|---|---|
| LiDAR | Livox Avia | 資料集感測器 | FAST-LIVO dataset | 240,000 points/s; mechanical, non-repetitive; 3 m to 450 m; FoV 70.4 x 77.2 deg | (Hong et al., 2024, Table I) |
| LiDAR | Ouster OS1-128 | 資料集感測器 | FusionPortable | 2,621,440 points/s; mechanical, repetitive; 1 m to 120 m; FoV 45 x 360 deg | (Hong et al., 2024, Table I) |
| LiDAR | RealSense L515歸入:Intel RealSense L515 | 方法輸入 | 未標示 | solid-state; 23,000,000 points/s and 9 m to 25 m as written; FoV 70 x 55 deg; indoor only | (Hong et al., 2024, Table I (Our Device I); Sec. IV-B) |
| LiDAR | Livox Avia | 方法輸入 | 未標示 | 240,000 points/s; mechanical, non-repetitive; 3 m to 450 m; FoV 70.4 x 77.2 deg; hardware-synchronized with camera | (Hong et al., 2024, Table I (Our Device II); Sec. III) |
| 慣性量測單元(IMU) | BM1088 (as written)歸入:BM1088 | 資料集感測器 | FAST-LIVO dataset | 原文未報告 | (Hong et al., 2024, Table I) |
| 慣性量測單元(IMU) | ICM20948 | 資料集感測器 | FusionPortable | 原文未報告 | (Hong et al., 2024, Table I) |
| 慣性量測單元(IMU) | BMI085 | 方法輸入 | 未標示 | 原文未報告 | (Hong et al., 2024, Table I (Our Device I)) |
| 慣性量測單元(IMU) | BMI088 | 方法輸入 | 未標示 | 原文未報告 | (Hong et al., 2024, Table I (Our Device II)) |
| 相機 | MV-CA013-21UC | 資料集感測器 | FAST-LIVO dataset | global shutter, 1280 x 1024, FoV 72 x 60 deg, hardware-synchronized | (Hong et al., 2024, Table I) |
| 相機 | BFS-U3-31S4C (written 'FILR BFS-U3-31S4C') | 資料集感測器 | FusionPortable | global shutter, 1024 x 768, FoV 66.5 x 82.9 deg | (Hong et al., 2024, Table I) |
| 相機 | RealSense L515 (RGB camera)歸入:Intel RealSense L515 | 方法輸入 | 未標示 | rolling shutter, 1920 x 1080, FoV 70 x 43 deg | (Hong et al., 2024, Table I (Our Device I)) |
| 相機 | MV-CA013-21UC | 方法輸入 | 未標示 | global shutter, 1280 x 1024, FoV 72 x 60 deg | (Hong et al., 2024, Table I (Our Device II)) |
| 運算硬體 | Intel Core i9 12900K | 執行運算平台 | 未標示 | 3.50 GHz | (Hong et al., 2024, Sec. IV) |
| 運算硬體 | NVIDIA GeForce RTX 4090 | 執行運算平台 | 未標示 | single GPU | (Hong et al., 2024, Sec. IV) |
| 其他 | ground-truth structure point clouds | 參考或真值量測 | FusionPortable; self-collected | provided for FusionPortable and both self-collected devices (acquisition instrument not stated) | (Hong et al., 2024, Table I; Sec. IV) |
作者報告的優勢與限制
優勢
- LiDAR-based initialization clearly improves CD, EMD and F-score over purely visual approaches (Sec. IV-C)
- Supports solid-state and mechanical LiDARs (abstract)
- Best interpolation PSNR (32.787) and extrapolation PSNR (19.220) in Table II; extrapolation PSNR 4.1 dB above 3D-GS (15.111) (Table II)
- On FusionPortable, full-model CD 0.107 and EMD 0.435 vs 0.149 and 0.698 for 3D-GS (Table IV)
- Full model (Case IV) exceeds 3D-GS interpolation and extrapolation PSNR on the average of five sequences (30.877 vs 30.168; 22.403 vs 21.185) (Table III)
限制
- Photometric optimization of Gaussian structure slightly reduced structural quality; pose refinement had mixed effects on F-score (Sec. IV-C)
- Training and rendering slower than 3DGS due to dense LiDAR points: 14m25s and 43 FPS vs 8m11s and 131 FPS (Sec. V, Table II)
- LiDAR structure is unreliable on glass and in over- or under-scanned areas, requiring photometric densification and pruning (Sec. III-C)
- LiDAR initialization without visual structure optimization (Case II) lowered interpolation PSNR relative to 3D-GS on four of five sequences (Sec. IV-B, Table III)
- F-score is highest for Case II (0.807) rather than the full model (0.751); CD and EMD units are not stated (Table IV)
營建工程相關證據
論文未在營建場域測試;資料為 FusionPortable、FAST-LIVO 及作者自錄室內外資料;作者提及數位孿生潛力但未驗證工程任務。
原文驗證環境:公開基準、受控實驗
報告的性能數據
以下是原文作者報告的性能數值(author-reported results),不是本研究重新量測的結果。每張圖只並列同一個比較組(comparison group,同一張表、同一組實驗設定)內的方法;不同比較組之間的數值不可直接比較,也不構成排名。
本方法共出現在 4 個比較組,合計 77 筆紀錄。
Shi et al., 2026 · Table I 本方法 48 筆
表格設定(擷取紀錄原文):Thermal Gaussian-splatting reconstruction quality; private scenes captured with the authors' Livox Avia + thermal rig at different times of day (Fig. 3 caption lists 12:00 p.m. to 6:00 a.m.), public scenes are five M2DGR sequences; EMD reference point cloud source not described; Avg. columns printed by the authors (Shi et al., 2026, Table I)
PSNR (rendering fidelity, higher is better),authors' private thermal dataset · Car
只並列這張表在相同設定下報告的方法;以「本方法:」開頭者為本頁方法。失敗、未執行與未報告以標記呈現,不是 0。
按 Tab 進入圖表後,用上下方向鍵逐一瀏覽各類別,Esc 關閉提示框;也可開啟表格檢視閱讀全部數值。
這些是 Shi et al., 2026 在此表設定下報告的數值(author-reported results),只能在同一個比較組內對照,不代表方法在其他資料或設定下的表現。
資料來源作者報告值(Shi et al., 2026, Table I)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Ours原文提出 | 26.061 | (Shi et al., 2026, Table I) |
| Thermal3D-GS | 24.216 | (Shi et al., 2026, Table I) |
| LIV-GaussMap本方法 | 23.569 | (Shi et al., 2026, Table I) |
Hong et al., 2024 · Table III 本方法 18 筆
表格設定(擷取紀錄原文):Ablation of map structure optimization: Case I = 3D-GS baseline; Case II = LiDAR-initialized Gaussians without visual structure optimization; Case III = Case II plus photometric position optimization; Case IV = full method with Gaussian pose refinement; SSIM and LPIPS rows of this table omitted here; Cases II and III are ablation variants (method_id null) (Hong et al., 2024, Table III)
PSNR [dB] (Interpolated),FAST-LIVO dataset · HKU_MB(outdoor)
只並列這張表在相同設定下報告的方法;以「本方法:」開頭者為本頁方法。失敗、未執行與未報告以標記呈現,不是 0。
按 Tab 進入圖表後,用上下方向鍵逐一瀏覽各類別,Esc 關閉提示框;也可開啟表格檢視閱讀全部數值。
這些是 Hong et al., 2024 在此表設定下報告的數值(author-reported results),只能在同一個比較組內對照,不代表方法在其他資料或設定下的表現。
資料來源作者報告值(Hong et al., 2024, Table III)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Case I (3D-GS baseline) | 24.39 dB | (Hong et al., 2024, Table III) |
| Case II (LiDAR initialization only) | 24.341 dB | (Hong et al., 2024, Table III) |
| Case III (+ photometric position optimization) | 24.24 dB | (Hong et al., 2024, Table III) |
| Case IV (full method)本方法原文提出 | 25.14 dB | (Hong et al., 2024, Table III) |
Hong et al., 2024 · Table II 本方法 8 筆
資料集與序列not stated (real-world dataset) · not stated
表格設定(擷取紀錄原文):Novel-view synthesis on interpolated and extrapolated views on a real-world dataset (dataset and sequence not named for this table); asterisk methods were enhanced with dense LiDAR point clouds; cost time and FPS on the authors' desktop (Hong et al., 2024, Table II)
PSNR (Interpolate),not stated (real-world dataset) · not stated
只並列這張表在相同設定下報告的方法;以「本方法:」開頭者為本頁方法。失敗、未執行與未報告以標記呈現,不是 0。
按 Tab 進入圖表後,用上下方向鍵逐一瀏覽各類別,Esc 關閉提示框;也可開啟表格檢視閱讀全部數值。
這些是 Hong et al., 2024 在此表設定下報告的數值(author-reported results),只能在同一個比較組內對照,不代表方法在其他資料或設定下的表現。
資料來源作者報告值(Hong et al., 2024, Table II)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Point-NeRF* [2] | 27.331 dB | (Hong et al., 2024, Table II) |
| DS-NeRF* [24] | 27.178 dB | (Hong et al., 2024, Table II) |
| 3D-GS* [1] | 31.9 dB | (Hong et al., 2024, Table II) |
| Plenoxel [18] | 26.744 dB | (Hong et al., 2024, Table II) |
| M-NeRF360 [16] | 28.446 dB | (Hong et al., 2024, Table II) |
| F2-NeRF [25] | 32.556 dB | (Hong et al., 2024, Table II) |
| 3D-GS [1] | 31.899 dB | (Hong et al., 2024, Table II) |
| Our method本方法原文提出 | 32.787 dB | (Hong et al., 2024, Table II) |
Hong et al., 2024 · Table IV 本方法 3 筆
資料集與序列FusionPortable · HKUST_indoor (per Table I)
表格設定(擷取紀錄原文):Structure accuracy of the Gaussian map against the ground-truth point cloud on FusionPortable (sequence HKUST_indoor per Table I); CD and EMD units not stated (Hong et al., 2024, Table IV)
CD (Chamfer Discrepancy [26]),FusionPortable · HKUST_indoor (per Table I)
只並列這張表在相同設定下報告的方法;以「本方法:」開頭者為本頁方法。失敗、未執行與未報告以標記呈現,不是 0。
按 Tab 進入圖表後,用上下方向鍵逐一瀏覽各類別,Esc 關閉提示框;也可開啟表格檢視閱讀全部數值。
這些是 Hong et al., 2024 在此表設定下報告的數值(author-reported results),只能在同一個比較組內對照,不代表方法在其他資料或設定下的表現。
資料來源作者報告值(Hong et al., 2024, Table IV)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Case I (3D-GS baseline) | 0.149 | (Hong et al., 2024, Table IV) |
| Case II (LiDAR initialization only) | 0.114 | (Hong et al., 2024, Table IV) |
| Case III (+ photometric position optimization) | 0.109 | (Hong et al., 2024, Table IV) |
| Case IV (full method)本方法原文提出 | 0.107 | (Hong et al., 2024, Table IV) |
來源
Hong et al., 2024
(2024)LIV-GaussMap: LiDAR-Inertial-Visual Fusion for Real-Time 3D Radiance Field Map RenderingIEEE Robotics and Automation Letters, 9(11), 9765-9772
DOI 10.1109/lra.2024.3400149arXiv 2401.14857程式碼
同儕審查已出版已讀全文近十年查證後修正
相關版本
- 預印本:arXiv:2401.14857 https://arxiv.org/abs/2401.14857
程式碼:https://github.com/sheng00125/LIV-GaussMap(授權:none (repository contains only README.md and figures; no source code or LICENSE at check time))。有公開程式碼不等於已被重現,也不代表目前版本與論文版本相同。