Though still a proof-of-concept, the underlying premise behind TensorView has been conclusively demonstrated using the search for abandoned buildings in an urban area as a testbed. There is, however, much work yet to be done. Most notably, the survey area preparation and manual review steps can be automated and streamlined, further reducing human workload and time commitment.
It is also necessary to begin to develop an "ecosystem" around TensorView. Most notably, it is (at present) impossible to distribute training sets without running afoul of international copyright law, unless the practitioner has created every image in the training set themselves. This necessitates the development of a standard-format "setfile" - essentially a text-based repository of images and their categories that can be retrieved over the Internet. Not only is distributing a text file much simpler than distributing a directory of images, it also does not violate copyright law, as the analyst is not distributing the images themselves - merely a list of locations where they may be acquired.
What is clear, however, is that the approach underlying TensorView - the automatic acquisition and analysis of Street View imagery - shows promise. The study conducted using this tool required approximately the same amount of time as a traditional "boots-on-the-ground" field survey but needed substantially less human effort, leaving the analyst free to work on other tasks. With further automation, this achievement can only be improved, and Street View can become an invaluable element of the analyst's toolkit - a possibility that creates a substantial number of new applications for GIS and remote sensing.