The stereo / flow benchmark consists of 194 training image pairs and 195 test image pairs, saved in loss less png format. Our evaluation server computes the average number of bad pixels for all non-occluded or occluded (=all groundtruth) pixels. We require that all methods use the same parameter set for all test pairs. Our development kit provides details about the data format as well as MATLAB / C++ utility functions for reading and writing disparity maps and flow fields.

Our evaluation table ranks all methods according to the number of non-occluded erroneous pixels at the specified disparity / end-point error threshold. All methods providing less than 100 % density have beeninterpolated using simple background interpolation as explained in the corresponding header file in the development kit. For each method we show:

• Out-Noc: Percentage of erroneous pixels innon-occluded areas
• Out-All: Percentage of erroneous pixels in total
• Avg-Noc: Average disparity / end-point error in non-occluded areas
• Avg-All: Average disparity / end-point error in total
• Density: Percentage of pixels for which ground truth has been provided by the method

Note: On 04.11.2013 we have improved theground truth disparity maps andflow fields leading to slightly improvements for all methods. Please download the stereo/flow dataset with theimproved ground truth for training again, if you have downloaded the dataset prior to 04.11.2013. Please consider reporting these new number for all future submissions. Links to last leaderboards排行榜 before the updates: stereo and flow!

•  Flow: Method uses optical flow (2 temporally adjacent images)
•  Multiview: Method uses more than 2 temporally adjacent images
•  Motion stereo: Method usesepipolar geometry 核面幾何學for computing optical flow
•  Additional training data: Use of additional data sources for training (see details)

Method	Setting	Code	Out-Noc	Out-All	Avg-Noc	Avg-All	Density	Runtime	Environment
1	1.77 %	2.30 %	0.6 px	0.7 px	100.00 %	0.9 s	Nvidia GTX Titan X
A. Kendall, H. Martirosyan, S. Dasgupta, P. Henry, R. Kennedy, A. Bachrach and A. Bry: End-to-End Learning of Geometry and Context for Deep Stereo Regression. arXiv preprint arxiv:1703.04309 2017.
2	code	2.27 %	3.40 %	0.7 px	1.0 px	100.00 %	48 s	Titan X (Torch7, CUDA)
3	2.28 %	3.48 %	0.7 px	0.9 px	100.00 %	71 s	TESLA K40C
4	SN	2.29 %	3.50 %	0.7 px	0.9 px	100.00 %	67 s	Titan X
5	PBCP	2.36 %	3.45 %	0.7 px	0.9 px	100.00 %	68 s	Nvidia GTX Titan X
6	code	2.37 %	3.09 %	0.7 px	0.8 px	100.00 %	265 s	>8 cores @ 3.0 Ghz (Matlab + C/C++)
7	code	2.43 %	3.63 %	0.7 px	0.9 px	100.00 %	67 s	Nvidia GTX Titan X (CUDA, Lua/Torch7)
8		code	2.46 %	2.69 %	0.8 px	0.8 px	99.93 %	70 s	GPU (Matlab + CUDA)
V. Ntouskos and F. Pirri: Confidence driven TGV fusion. arXiv preprint arXiv:1603.09302 2016.
9	code	2.47 %	3.27 %	0.7 px	0.9 px	100.00 %	265 s	>8 cores @ 3.0 Ghz (Matlab + C/C++)
10	2.61 %	3.84 %	0.8 px	1.0 px	100.00 %	100 s	Nvidia GTX Titan (CUDA, Lua/Torch7)
11	PRSM		code	2.78 %	3.00 %	0.7 px	0.7 px	100.00 %	300 s	1 core @ 2.5 Ghz (C/C++)
12		2.83 %	3.64 %	0.8 px	0.9 px	100.00 %	35 s	1 core @ 3.5 Ghz (C/C++)
13	3.02 %	4.45 %	0.8 px	1.0 px	100.00 %	1.35 s	1 core 2.5 Ghz + K40 NVIDIA, Lua-Torch
14	VC-SF		3.05 %	3.31 %	0.8 px	0.8 px	100.00 %	300 s	1 core @ 2.5 Ghz (C/C++)
C. Vogel, S. Roth and K. Schindler: View-Consistent 3D Scene Flow Estimation over Multiple Frames. Proceedings of European Conference on Computer Vision. Lecture Notes in, Computer Science 2014.
15	3.07 %	4.29 %	0.8 px	1.0 px	100.00 %	0.7 s	Nvidia GTX Titan X (Torch)
16	TBR	3.09 %	4.29 %	0.7 px	0.9 px	100.00 %	1750 s	4 cores @ 3.0 Ghz (Matlab + C/C++)
17	3.10 %	4.24 %	0.9 px	1.1 px	100.00 %	3 s	1 core @ 2.5 Ghz (C/C++)
18	JSOSM	3.15 %	3.94 %	0.8 px	0.9 px	100.00 %	105 s	8 cores @ 2.5 Ghz (C/C++)
19	LPU	3.22 %	4.27 %	0.8 px	1.0 px	100.00 %	1650 s	1 core @ 2.5 Ghz (C/C++)
20	OSF		code	3.28 %	4.07 %	0.8 px	0.9 px	99.98 %	50 min	1 core @ 3.0 Ghz (Matlab + C/C++)
M. Menze and A. Geiger: Object Scene Flow for Autonomous Vehicles. Conference on Computer Vision and Pattern Recognition (CVPR) 2015.
21	CoR	code	3.30 %	4.10 %	0.8 px	0.9 px	100.00 %	6 s	6 cores @ 3.3 Ghz (Matlab + C/C++)
22	3.32 %	5.24 %	0.9 px	1.9 px	100.00 %	60 s	4 cores @ 3.5 Ghz (C/C++)
23	code	3.39 %	4.41 %	0.9 px	1.0 px	100.00 %	2 s	1 core @ 3.5 Ghz (C/C++)
24	3.40 %	4.72 %	0.8 px	1.0 px	100.00 %	5 min	4 cores @ 2.5 Ghz (Matlab + C/C++)
25	STD	3.48 %	4.24 %	0.9 px	1.0 px	99.99 %	40 min	2 cores @ 3.0 Ghz (C/C++)
26	MN	3.58 %	4.77 %	0.9 px	1.1 px	100.00 %	3 min	>8 cores @ 2.5 Ghz (C/C++)
27	CPM2	code	3.58 %	4.41 %	0.9 px	1.1 px	99.99 %	1.8 s	1 core @ 2.5 Ghz (C/C++)
28	3.83 %	4.59 %	0.9 px	1.0 px	100.00 %	1 min	1 core @ 2.5 Ghz (Matlab + C/C++)
D. Wei, C. Liu and W. Freeman: A Data-driven Regularization Model for Stereo and Flow. 3DTV-Conference, 2014 International Conference on 2014.
29	3.92 %	5.11 %	0.9 px	1.0 px	99.89 %	2.3 s	1 core @ 3.0 Ghz (C/C++)
30	SMCM	3.94 %	5.24 %	0.9 px	1.1 px	100.00 %	1800 s	Nvidia GTX 1080 (Caffe)
31		4.02 %	4.87 %	0.9 px	1.0 px	100.00 %	200 s	4 cores @ 3.0 Ghz (Matlab + C/C++)
C. Vogel, K. Schindler and S. Roth: Piecewise Rigid Scene Flow. International Conference on Computer Vision (ICCV) 2013.
32	PCBP	4.04 %	5.37 %	0.9 px	1.1 px	100.00 %	5 min	4 cores @ 2.5 Ghz (Matlab + C/C++)
33	code	4.11 %	4.65 %	0.9 px	1.0 px	100.00 %	0.06 s	Nvidia GTX Titan X (Caffe)
34	CSPMS	4.13 %	5.92 %	1.2 px	1.6 px	100.00 %	6 s	4 cores @ 2.5 Ghz (C/C++)
35	4.27 %	5.33 %	1.0 px	1.1 px	100.00 %	5 s	4 cores @ 2.5 Ghz (C/C++)
36	4.27 %	5.45 %	1.0 px	1.1 px	100.00 %	2 s	4 cores @ >3.5 Ghz (C/C++)
37	MBM	4.35 %	5.43 %	1.0 px	1.1 px	100.00 %	0.2 s	1 core @ 3.0 Ghz (C/C++)
38		4.36 %	5.22 %	0.9 px	1.1 px	100.00 %	150 sec	4 core @ 3.0 Ghz (Matlab - C/C++)
C. Vogel, K. Schindler and S. Roth: Piecewise Rigid Scene Flow. International Conference on Computer Vision (ICCV) 2013.
39	code	4.49 %	5.26 %	1.0 px	1.2 px	96.37 %	6 s	6 cores @ 3.3 Ghz (Matlab + C/C++)
40	sgm	4.50 %	5.74 %	1.1 px	1.3 px	96.89 %	1 s	1 core @ 2.5 Ghz (C/C++)
41	TVTGV	4.60 %	6.09 %	1.3 px	1.5 px	100.00 %	5.6 s	GPU @ 2.5 Ghz (C/C++)
42	AEGF	4.81 %	6.12 %	1.2 px	1.8 px	99.99 %	8 s	1 core @ 2.5 Ghz (C/C++)
43	AARBM	4.86 %	5.94 %	1.0 px	1.2 px	100.00 %	0.25 s	1 core @ 3.0 Ghz (C/C++)
44	wSGM	4.97 %	6.18 %	1.3 px	1.6 px	97.03 %	6s	1 core @ 3.5 Ghz (C/C++)
45	AABM	4.97 %	相關推薦 The KITTI Vision Benchmark Suite之Stereo Evaluation 2012 The stereo / flow benchmark consists of 194 training image pairs and 195 test image pairs, saved in loss less png format. Our evaluation server compu Cracking the coding interview 智力題之-扔雞蛋問題 ack TP int inter eight com 技術 png nbsp 問題： 解法： 這個問題算是一個智力題，沒看過相關題目很難第一時間想出來。其中的“最差情況平衡”思路值得學習。 Cracking the coding interview 智力題之-扔雞蛋問 Burp Suite之截斷代理功能及相關設定（一） Burpsuite 1.burpsuite 簡介2.設定代理3.Target4.爬網模組5.掃描模組6.擴充套件模組7.intrude8.Repeater9.Sequencer10.Decoder11.Compare12.burpsuite 專案主配置資訊     1.burpsu Extending the Yahoo! Streaming Benchmark could accomplish with Flink back at Twitter. I had an application in mind that I knew I could make more efficient by a huge factor if I could use the Kitti資料集使用之坑 KITTI資料集簡介與使用： https://blog.csdn.net/solomon1558/article/details/70173223   接下來是我使用時遇到的坑： 一、label檔案中ry的含義 按照文件中圖示，ry應該是小車朝向和相機座標系z軸的夾角 The Tech Vision Problem I was once fascinated by new technology.It has been painful for me to write about the tech industry, because as a kid, I was drawn to the very aspects of t The Minimum Viable Test Suite In the last post we detailed how to validate email addresses during user registration. This time, we’ll add unit and integration tests (yay!) to our app Middleburry stereo Evaluation示例資料的執行及深度圖結果檢視 先丟出來Middleburry stereo Evaluation的網址吧：已經到了第3版了，最新的資料集是2014年的跑通用的是ubuntu17.10（之前本科的時候，用windows跑，各種難過，ubuntu下也有問題，不過相對來講，容易多了，哈哈哈哈）。跑通這個東東，還 stereo evaluation 感謝前輩們的分享~ 在安裝cvkit時，執行cmake..出錯：-- Could NOT find GLUT (missing:  GLUT_glut_LIBRARY GLUT_INCLUDE_DIR)  解決方法：執行下面這兩句 sudo apt- VMware Workstation 11 搭建windows server 2012 之sql server 2012集群常見問題整理 額外 stat 獲得 系統 網絡 net .com scsi 支持 1、windows server 2012內置支持iSCSI發起程序無需額外安裝，iSCSI Software Target 可作為“文件和存儲服務”角色下的內置功能使用 2 微電子新手入門之Windows Server 2012 R2 Standard多使用者遠端控制 注意：配置新伺服器時需要如此操作，正常連線使用伺服器不需要以下步驟，設定完成後，即可實現由單使用者遠端控制改成多使用者遠端控制。 說明：Windows Server 2012預設情況下，只執行2個使用者遠端桌面登陸，這裡我們可以通過安裝遠端桌面會話主機配置來實現2個以上使用者的遠端桌面登陸。 數據可視化入門之show me the numbers 推薦 有趣的 好的 style blank 分享 span 需要 width 數據的可視化一直是自己瞎玩著學，近來想系統的學數據可視化的東西，於是搜索資料時看到有人推薦《show me the numbers》作為入門。 由於搜不到具體的書籍內容，只能搜到一個 tomcat server 報錯之 More than the maximum allowed number of cookies int str sock lar time color errors round pin More than the maximum allowed number of cookies EVERE: Error processing request java.lang. SVN 問題解決之 The XML response contains invalid XML 根目錄 項目目錄 XML svn fix 目錄 val 執行 操作 公司幾個同事的SVN更新時出現了The XML response contains invalid XML報錯 經Google得到一個線索,可能和Http請求有關. 想起之前項目改過一次網絡請求方式,把 Android錯誤之--Error retrieving parent for item: No resource found that matches the given name &#39;Theme.A 方法 minimum vertical 宋體 popu pac fim wid bre 錯誤提示：error: Error retrieving parent for item: No resource found that matches the given name Vue入門之旅：一報錯 Unknown ... make sure to provide the "name" option及error compiling template ont methods rect return tex exactly gist () 編譯 報錯一： Unknown custom element: <custom-select> - did you register the component correc 每日一水之 luogu2907 [USACO08OPEN]農場周圍的道路Roads Around The Farm tin etime 輸入輸出 empty arm around class using top 題目描述 Farmer John‘s cows have taken an interest in exploring the territory around the far what's the python之基本運算符及字符串、列表、元祖、集合、字典的內置方法 back 原理 pop mat 要點 log 程序 eric none 　　計算機可以進行的運算有很多種，運算按種類可分為算數運算、比較運算、邏輯運算、賦值運算、成員運算、身份運算。字符串和列表的算數運算只能用+和*，字典沒有順序，所以不能進行算數運算和比較運算。比較運算中 what' the python之遞歸函數與二分算法 hat 簡單的 not post col 組成 成了 n) spa what‘s the 遞歸？ 　　遞歸函數的定義：在函數裏可以再調用函數，如果這個調用的函數是函數本身，那麽就形成了一個遞歸函數。 　　遞歸的最大深度為997，這個是程序強制定義的，997完全可以滿足一般情 what's the python之函數及裝飾器 聲明 做的 運用 本質 修改 忽略 局部變量 即使 實現 what‘s the 函數？ 　　函數的定義：（return是返回值，可以沒有，不過沒有的話就返回了None） def wrapper(參數1,參數2,*args,默認參數,**kwargs): ‘‘‘註釋