Information Technology Laboratory,縮稱ITL,是NIST七個研究實驗室之一,涵蓋計算機科學,數學,統計學和系統工程等領域。根據網路的文章,ITL實驗室有將近500名研究人員,年預算額近1.5億美元。
ITL底下再細分為七個研究部門
計算機安全部(Computer Security Divsion, CSD)
先進網絡技術部(Advanced Network Technologies Division)
應用和計算數學部( Applied and Computational Mathematics Division)
應用資安部(Applied Cybersecurity Division)
資訊擷取部( Information Access Division) 底下再分為 圖像組(Image Group)、多模式信息組(Multimodal Information Group)、檢索組(Retrieval Group)、可視化和可用性組(Visualization and Usability Group)
FRVT is an ongoing activity, and all evaluations run continuously with no submission deadlines. For the FRVT 1:1, 1:N, and Quality tracks, participants may send ONE submission as often as every four calendar months from the last submission for evaluation. For FRVT MORPH, the number and schedule of submissions is currently not limited, so participants can send submissions at any time. Algorithm submissions will be processed on a first-come first-serve basis for inclusion in subsequent reports.
Missing entries for visa, mugshot and wild images generally mean the algorithm did not run to completion. For child exploitation, missing entries arise because NIST executes those runs only infrequently
The algorithms are ordered in terms of lowest mean rank across mugshot, visa, visa border, and wild datasets, rewarding broad accuracy over a good result on one particular dataset.
這個排行將mugshot, visa, visa border, and wild等資料集的測驗結果排名平均,也就是說,你不一定要每一項都是最好的,你只要整體平均好就好。
The evaluation used four datasets — frontal mugshots, profile views, webcam photos and wild images — and the report lists accuracy results alongside developer names
mugshot、profile、webcam、wild是主要的四個資料集
The primary dataset is comprised of 26.6 million reasonably wellcontrolled live portrait photos of 12.3 million individuals.
資料集的數量大概有2.6千萬張照片、1.2千萬個人
From Fall 2019 this report will be updated continuously as new algorithms are submitted to FRVT, and run on new datasets. Participation in the one-to-many identification track requires a devloper to first demonstrate high accuracy in the one-to-one verification track of FRVT.
In the same way that we’ve seen the science of genetically modified foods get set back 10 years because the public in Europe wouldn’t support it, we need to find a path which gets social license to operate.
舉個例子,假設我是 A 公司,我們公司是做購物網站的,而通常金流這一塊並不會自己做,而是會找其他做金流的廠商合作,在後端去「串接」金流服務商提供的功能,講白話一點就是:「當使用者要付款時,我把使用者導過去金流廠商的頁面,付款完再導回來我們網站」,相信有在網路上購物的大家應該很熟悉這個流程。
在這個過程中,雙方都必須留下紀錄,確保未來發生問題時有證據可以輔助說明。
例如說有天 A 公司突然接到一堆客訴說沒辦法付款,這時 A 公司直接打電話去金流商,罵說你們這什麼爛服務,怎麼突然壞掉,而金流商此時提供了伺服器的 log,說:「沒有啊,我們這邊從今天早上八點開始就沒有你們導過來的紀錄了,應該是你們的問題吧?」,後來 A 公司檢查了自己這邊的服務,確實是因為今天早上的版本更新出了問題而導致,跟金流商一點關係都沒有。