For media librarians, the useful question in Google’s March 8, 2017 announcement was who would do each part of a search. Google introduced Cloud Video Intelligence API in private beta, describing searches that could locate shots containing particular entities in videos held in Google Cloud Storage. It named media catalogs among the intended uses. That offered a trial to design, not evidence that archive staff had become unnecessary. [video]
Our proposed workflow gives the machine the first retrieval pass and a librarian the next judgment. Staff would choose a small collection, compare returned shots with familiar footage and record mistaken or missed matches. A catalog maintainer would then decide which descriptions to retain. These are suggested responsibilities for testing Google’s stated capability, not steps the company proved would save time. [video]
A separate announcement said Kaggle would join Google Cloud. Google described its community as working with public datasets and building machine-learning models, and promised continued training and deployment support alongside dataset storage and querying. For an archive manager, that is a separate skills question: exploring model development is not the same assignment as checking whether a retrieved clip belongs in a catalog. [kaggle]
A useful trial record would therefore name the retrieval operator, the clip checker and the catalog owner, with time spent on each task recorded rather than assumed. Keep community learning separate from the operational test. The two Google posts support those different questions but do not independently establish performance. A decision about staffing would need evidence from the archive’s own work, beyond the announcement. [video][kaggle]
Source documents
- Google Cloudhttps://cloud.google.com/blog/products/gcp/announcing-google-cloud-video-intelligence-api-and-more-cloud-machine-learning-updates
- Google Cloudhttps://cloudplatform.googleblog.com/2017/03/welcome-Kaggle-to-Google-Cloud.html