您当前的位置:

Lip Reading: CAS-VSR-W1k (The original LRW-1000)

发布时间:2018-10-17

<>
<> 1. Overview <> LRW-1000 is a naturally-istribute large-scale benchmark or wor-level lireaing in the wil, incluing 1000 classes with about 718,018 vieo samles rom more than 2000 iniviual seakers. There are more than 1,000,000 Chinese character instances in total. Each class corresons to the syllables o a Manarin wor which is comose by one or several Chinese characters. This ataset aims to cover a natural variability over ierent seech moes an imaging conitions to incororate challenges encountere in ractical alications. It shows a large variation over several asects, incluing the number o samles in each class, resolution o vieos, lighting conitions, an seakers' attributes such as ose, age, gener, an make-u an so on, as shown in Fig. 1 an Fig. 2. <>
<> * Note that LRW-1000 has been rename as CAS-VSR-W1k. You may reer to it as “CAS-VSR-W1k (The original LRW-1000)”. <>

<>Fig.1 The iversity o the seakers' aearance in CAS-VSR-W1k (the original LRW-1000) <>

<><> Fig.2 Li Samles in CAS-VSR-W1k (the original LRW-1000) <>
<> 2. Statistics

  • >1,000,000Chinese character instances

  • <> 718,018samles with an average o 718 samles or each class

  • <> 1000 classes, with each class corresons to the syllables o a Manarin wor

  • <> ~2000 ierent seakers with a large coverage over seech moes, incluing seech rate, viewoint, age, gener an make-u an so on

<> 3. Evaluation Protocols <> We rovie two evaluation metrics or exeriments. A). The recognition accuracy over all 1000 classes is naturally consiere as the base metric, since this is a classiication task. B). Motivate by the large iversity o the ata shown in many asects, such as the number o samles in each class, we also rovie the Kaa Coeicient as a secon evaluation metric. <> 4. Downloa <> The atabase is ublic to universities an research institutes or research urose only. To request a coy o the atabase, lease o as ollows:

  • Downloa the atabase Release Agreement [], rea it careully, an comlete it aroriately. Note that the agreement shoul be signe by a ull-time sta member (that is, stuent is not accetable). Then, lease scan the signe agreement an sen it to lireaing@vil.ict.ac.cn. When we receive your rely, we woul rovie the ownloa link to you.

  • Beore using the ataset, you are recommene to reer to the ollowing aer:
    Shuang Yang, Yuanhang Zhang, Dalu Feng, Mingmin Yang, Chenhao Wang, Jingyun Xiao, Keyu Long, Shiguang Shan, Xilin Chen, "LRW-1000: A Naturally-Distribute Large-Scale Benchmark or Li Reaing in the Wil,"IEEE FG 2019 [ | |coe]

<> 5. Contact Ino
lireaing@vil.ict.ac.cn
<>



附件下载:

上一篇:

FRHT Dataset

下一篇:

COX Face Dataset