This document discusses using visual analysis techniques to access large audiovisual collections. It describes how digitization programs have created collections with over 800,000 hours of content. Automatic visual analysis techniques like face recognition, speech recognition and visual concept detection could help users find content in these large collections. The document outlines challenges like needing training examples to perform detection and managing user expectations about the capabilities and limitations of these techniques. It advocates exploring how these technologies could help with tasks like query by example for digital humanities researchers while being aware of limitations.
29. Expectation Management
• Expectation management:
– Training examples versus result list
– Google images search versus visual search in AV
• Understanding visual search:
– why something is hard to detect
• visual characteristics, training examples
– Noise is not bad per definition
30. DH perspective
• First explorations in various projects
– Requirements studies
– Demonstrations
– Prototypes
• Technology is ready to start exploring its use in
real use scenarios (e.g., query by example)
• Feed DH ideas into ICT research community
31. Technology exists that could help
Technology does not solve all
problems
Discuss with ICT experts
Technology has a price, what is the
RoI?
AWARENESS
How does technology fit in
How do limitations fit in
‘Technology Critique’ (Historian 2.0?)
ICT and curriculum
METHODOLOGY/TRAINING
What can it do?
How does it work?
How does it perform?
How can it be improved?
MICRO MACRO
How can we use it?
What do we need?
How does it scale?
Who could benefit as well?