Identifying Latent Attributes From Video Scenes Using Knowledge Acquired From Large Collections Of Text Documents