前幾天收信看到的多媒體資料庫 因為裡頭資訊量很大 研究有相關的人可以考慮抓(或寫信去要)一下 根據以往經驗 這種類型的資料庫 除非建構單位很佛心，願意永續經營 不然一兩年後，那些人若不在那個單位 這個資料要抓就再也抓不到了 -- http://research.microsoft.com/en-us/projects/msrammdata/ MSRA-MM 1.0 (Release Date: March 16 2009) MSRA-MM 1.0 consists of two sub-datasets, i.e., an image dataset and a video dataset that are collected from the image and video search engines. For image dataset, we have selected 68 representative queries based on the query log of search engies and then collect about 1000 images for each query. There are 65443 images in all. For video dataset, we have selected 165 representative queries from query log and collect 10277 videos accordingly. Features and annotations are provided. More details please read the technical report. Instruction.pdf ftp://[email protected]/pub/MSRAMM/Instruction.pdf Dataset(around 10G) ftp://[email protected]/pub/MSRAMM/MSRAMM_Release.zip.001 ftp://[email protected]/pub/MSRAMM/MSRAMM_Release.zip.002 ftp://[email protected]/pub/MSRAMM/MSRAMM_Release.zip.003 ※MSRA-MM 2.0 (Release Date: July 10 2009) The image part contains about 1 million images from 1165 queries and the video part contains 23 thousands of videos. The associated web pages are also downloaded and surrounding texts are extracted. More details please read the dataset description. --

※ 發信站: 批踢踢實業坊(ptt.cc) ◆ From: 122.116.111.182