Deep Fakes for Text

I attended a talk recently where they discussed how the capacity to produce deep faked video would affect our capacity to trust what we’re seeing as evidence in the media and in the court systems.  At that discussion, some mentioned that perhaps the textual media would then retain some form of advantage in the credibility game…


Researchers, scared by their own work, hold back “deepfakes for text” AI


Automated meeting notes

So – would you let somebody listen in to all your meetings to stop having to take your own notes?

voicea – Turn talk into action

What forms of informal or formal permissions will we need to establish as we start having automated systems that could be listening in to our every conversation and producing notes?

Autonomous Delivery

George Mason is sharing the details of an automated food delivery service on campus.

A fleet of Starship Robots

These are the same bots that had at one point been slated to be navigating Austin’s streets.  Good to see Starship Technologies still plugging away at the problem.

Still curious how we could use the work in this area to change the way libraries transport our volumes around our campuses.

update 2019-02-04:

Adding a link to Amazon Scout:

Meet Scout: Field testing a new delivery system with Amazon Scout.

And also to LG CLOI:

A robot that actively engages with people, providing information
and services such as escort, ordering, delivery, shopping and cleaning.

Reducing Noise in Digital Images

NVidia has announced impressive progress in using AI to remove noise from “grainy” images without access to a clean version of the image to learn from.

By noise – they tend to refer to the grainy result of a low light digital photo, a side benefit being that they can also easily remove textual noise.  Currently, the result is “softer” than the original clean image, but I’m curious whether it will end up causing issues with watermarking or other copy protection schemes.  At what point will “good enough” be sufficient for a derivative use when we deal in low resolution imagery on the web all the time?

Many of us in collections rely on the use of watermarks to make openly sharing our collections more palatable to our donors.  Already, we have to warn them that there is no low barrier way to really prevent unattributed image reuse… This is simply going to make that conversation even more difficult.


Detecting Book Spines in an Image

I have images of stacks of books – can I detect the spines? Can I get the call numbers?

Combining Image and Text Features: A Hybrid Approach to Mobile Book Spine Recognition
(combine text recognition with comparing to known images of book spines)

Automatic Book Spine Extraction and Recognition for Library Inventory Management

Discussion on the OPENCV forum

Matching book-spine images for library shelf-reading process automation

Smart Library: Identifying Books on Library Shelves Using Supervised Deep Learning for Scene Text Reading

Mobile augmented reality for books on a shelf

Viewpoint-independent book spine segmentation

Identifying books in library using line segment detector and contour clustering

A review of Augmented Reality and its application in context aware library system