Poisson Surface Reconstruction
作者指出定向點(oriented points)的法向量可視為實體指示函數(indicator function)梯度的取樣,於是把表面重建轉為泊松方程式求解,再擷取等值面成為封閉網格。解法一次考慮全部點,不需啟發式分區與混合,因而對雜訊具韌性;並在八元樹上以局部支撐基底形成稀疏且條件良好的線性系統,另處理非均勻取樣。
本頁內容
Casts reconstruction from oriented points as a Poisson problem for an indicator function solved globally on an adaptive octree, then extracts a watertight isosurface.
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 | 未記錄 |
|---|---|
| 原文測試平台 | 未記錄 |
| 狀態估計 | 不適用 |
| 資料關聯 | 不適用 |
| 時間表示 | 不適用 |
| 去畸變 | 不適用 |
| 迴圈閉合 | 不適用 |
| 全域最佳化 | global Poisson solve over all oriented points (octree, multiscale) |
| 地圖表示 | implicit indicator function on an adaptive octree |
| 先驗資訊 | oriented normals required |
| 可輸出幾何 | watertight triangle mesh (isosurface of the indicator function) |
| 計算需求 | offline (hardware and processor type not reported); conjugate-gradient solve per octree depth in a multigrid-like scheme with block Gauss-Seidel to cap memory (Sec. 4.3); time and memory in Tables 1-2 |
使用設備
尚未收錄此方法的設備紀錄;設備資料仍在分批查證,沒有紀錄不代表原文未使用任何設備。
作者報告的優勢與限制
優勢
- ["Global solution without heuristic partitioning, resilient to noise (Sec. 1).", "Sparse, well-conditioned system via locally supported octree basis (Sec. 1, Sec. 2).", "Recovers sharp creases where VRIP shows 'lipping' which the authors attribute to VRIP's distance function being grown perpendicular to the view direction rather than to the surface normal (Forma Urbis fragment and Happy Buddha, Sec. 5.2, Figs. 5-6).", "Time and memory roughly quadratic in resolution
- the David head at depth 11 (215,613,477 samples) took 1.9 h and 5.2 GB and produced 16,328,329 triangles (Sec. 5.3, Table 1).", "Adaptive filter width gives smoother fits than the fixed-resolution FFT method in sparsely sampled regions without losing detail elsewhere (Sec. 5.2, Fig. 7)."]
限制
- ["Does not use acquisition-modality information such as line of sight: with no samples between the Happy Buddha's feet the surface connects them, whereas VRIP carves them apart (Sec. 5.2 'Limitation of our approach', Fig. 6).", "On the Stanford Bunny (Poisson at depth 9) it was neither fastest nor most memory-efficient: 263 s and 310 MB, versus 28 s for the fastest method (MPU) and 186 MB for the most memory-efficient (VRIP) (Sec. 5.3, Table 2).", "Requires oriented normals
- for the bunny they were estimated from neighbouring positions (Sec. 5.2).", "Follow-up by the same group reports a tendency to over-smooth the data (Kazhdan and Hoppe 2013, Sec. 1).", "Follow-up LiDAR work reports that the watertight assumption extrapolates surfaces where no data exist, requiring density-based trimming (Vizzo et al. 2021, Sec. III-B)."]
營建工程相關證據
原論文以物件掃描資料(Stanford Bunny、Dragon、Happy Buddha、Forma Urbis Romae 殘片、David)示範,未涉及營建。作者自述不使用視線資訊,因此在無資料處會把分離的部件連起來(Fig. 6);加上封閉曲面假設,在開放的室外或施工場景需額外修剪(依 Vizzo et al. 2021)。
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 5 個比較組,合計 35 筆紀錄。以下列出本方法紀錄最多的 4 組,其餘 1 組列在最後,並連到性能比較頁。
Kazhdan & Hoppe, 2013 · Table I 本方法 16 筆
表格設定(擷取紀錄原文):Wall-clock time and memory for Neptune and David at depths 8 to 11; screening weight alpha = 4, Neumann boundaries, samples-per-node 1; bracketed values are the new solver with alpha = 0; dagger: SSD at David depth 11 reports CPU user time because memory exceeded RAM (Kazhdan & Hoppe, 2013, Table I)
Time in seconds,Neptune (Aim@Shape) · depth 8
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 Kazhdan & Hoppe, 2013 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(Kazhdan & Hoppe, 2013, Table I)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Poisson本方法硬體:laptop, quad-core Intel Core i7, 8 GB RAM | 10 s | (Kazhdan & Hoppe, 2013, Table I) |
| Wavelet硬體:laptop, quad-core Intel Core i7, 8 GB RAM | 3 s | (Kazhdan & Hoppe, 2013, Table I) |
| SSD硬體:laptop, quad-core Intel Core i7, 8 GB RAM | 275 s | (Kazhdan & Hoppe, 2013, Table I) |
| Screened原文提出硬體:laptop, quad-core Intel Core i7, 8 GB RAM | 14 s | (Kazhdan & Hoppe, 2013, Table I) |
Kazhdan et al., 2006 · Table 1 本方法 12 筆
表格設定(擷取紀錄原文):Dragon model reconstructed at octree depths 7 to 10; kernel depth 6 for density estimation; hardware not reported (Kazhdan et al., 2006, Table 1)
Time (s),Stanford dragon · depth 7
這張表在此指標與資料序列只列出本方法一筆,沒有可並列的其他方法,因此不畫圖,數值與出處見下表。這是 Kazhdan et al., 2006 在此表設定下報告的數值(author-reported results),不代表方法在其他資料或設定下的表現。
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Poisson本方法原文提出 | 6 s | (Kazhdan et al., 2006, Table 1) |
Kazhdan et al., 2006 · Table 2 本方法 3 筆
資料集與序列Stanford Bunny · raw data, 362,000 points (Poisson at depth 9)
表格設定(擷取紀錄原文):Stanford Bunny raw data (362,000 points from ten range images), processed to fit each algorithm's input format; Poisson reconstructed at octree depth 9, resolution settings of the other seven methods not reported (VRIP used the registered scans with confidence values); running time in seconds, peak memory in MB, output triangles; hardware not reported (Kazhdan et al., 2006, Table 2)
Time (s),Stanford Bunny · raw data, 362,000 points (Poisson at depth 9)
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 Kazhdan et al., 2006 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(Kazhdan et al., 2006, Table 2)
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Power Crust | 380 s | (Kazhdan et al., 2006, Table 2) |
| Robust Cocone | 892 s | (Kazhdan et al., 2006, Table 2) |
| FastRBF | 4919 s | (Kazhdan et al., 2006, Table 2) |
| MPU | 28 s | (Kazhdan et al., 2006, Table 2) |
| Hoppe et al 1992 | 70 s | (Kazhdan et al., 2006, Table 2) |
| VRIP | 86 s | (Kazhdan et al., 2006, Table 2) |
| FFT | 125 s | (Kazhdan et al., 2006, Table 2) |
| Poisson本方法原文提出 | 263 s | (Kazhdan et al., 2006, Table 2) |
Kazhdan et al., 2006 · Text Sec. 5.3 本方法 3 筆
資料集與序列David (non-rigidly aligned scans) · head, depth 11
表格設定(擷取紀錄原文):Head of Michelangelo's David at depth 11 from 215,613,477 samples (Kazhdan et al., 2006, Text Sec. 5.3)
computation time,David (non-rigidly aligned scans) · head, depth 11
這張表在此指標與資料序列只列出本方法一筆,沒有可並列的其他方法,因此不畫圖,數值與出處見下表。這是 Kazhdan et al., 2006 在此表設定下報告的數值(author-reported results),不代表方法在其他資料或設定下的表現。
| 方法(原文寫法) | 報告值 | 出處 |
|---|---|---|
| Poisson本方法原文提出 | 1.9 h | (Kazhdan et al., 2006, Sec. 5.3, Fig. 8) |
其他比較組
列出其餘 1 個比較組
來源
Kazhdan et al., 2006
(2006)Poisson Surface ReconstructionEurographics Symposium on Geometry Processing (SGP 2006), pp. 61-70
DOI 10.2312/sgp/sgp06/061-070程式碼
同儕審查已出版已讀全文經典查證後修正
相關版本
- 後續方法:Screened poisson surface reconstruction (Kazhdan and Hoppe, ACM TOG 2013), a later extension of the method by a subset of the authors, not a journal version of this paper 10.1145/2487228.2487237
程式碼:https://github.com/mkazhdan/PoissonRecon(授權:MIT (repository LICENSE; repository is the first author's and implements the later screened version))。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。