If "image captioning" is utilized to make a commercial product, what application fields will need this technique? And what is the level of required performance for this technique to be usable?
If "image captioning" is utilized to make a commercial product, what application fields will need this technique?
There are several important use case categories for image captioning, but most are components in larger systems, web traffic control strategies, SaaS, IaaS, IoT, and virtual reality systems, not as much for inclusion in downloadable applications or software sold as a product. These are a few examples.
- Provision of captions to reduce the number of user operations (keying and selection) required to post a movie or image to improve posting volume of sticky media and therefore improve the position of a web site or page in terms of human interest
- Provision of captions to provide HTML header and alt attribute content to improve search engine scoring of page for search terms related to the content of the movie or image
- Testing to see if the content of an image submitted for posting matches the criteria intended by the owner or stake holder for the posting space on the basis of caption rather than directly through CNN categorization
The first two are usually monetized in ways such as these.
- Improvements in online purchase daily volume
- Acquisition of contact lists with consumer or business interests as fields in the list for marketing purposes
Additional draw of web traffic to enhance ad impression revenue
What is the level of required performance can this technique be usable?
Performance is in terms of these system qualities.
- Cost of computing resources per caption generated
- Accuracy of caption generated (which may be quantified in more than one way)
- Reliability of generation (since the AI component should produce a rating of confidence with the caption string or produce a null or empty string based on a hyper-parameter confidence threshold
The quantification of cost, accuracy, and reliability is business dependent. Some may have a nasty negative effect on the business if the caption is wrong or missing. Others use cases may not.
In some cases the average revenue generated by the caption's presence is already known and known to be small, which requires that the CNN run time and computing resource requirements must be kept below that. In other cases, a Fortune 500 company CTO said, "Do it and send me the budget needed." In such cases the budget may be, for all practical purposes, unconstrained, not that a system that wastes time and resources is ever desirable, even if only for energy conservation reasons.