This README.txt file was generated on 2020-10-09 by Matthew Lincoln # # General instructions for completing README: # For sections that are non-applicable, mark as N/A (do not delete any sections). # Please leave all commented sections in README (do not delete any text). # ------------------- GENERAL INFORMATION ------------------- 1. Title of Dataset: CAMPI Screencasts Screen captures documenting several interfaces from the CAMPI project. Full project whitepaper: Lincoln, Matthew, Julia Corrin, Emily Davis, and Scott Weingart. "CAMPI: Computer-Aided Metadata Generation for Photo Archives Initiative." Pittsburgh: Carnegie Mellon University Libraries, October 8, 2020. https://doi.org/10.1184/R1/12791807. 2. Author Information First Author Contact Information Name: Matthew Lincoln Institution: Carnegie Mellon University Libraries Address: 5000 Forbes Avenue, Pittsburgh, PA 15213-3890 Email: mlincoln@andrew.cmu.edu ----------------------------------------- DATA DESCRIPTION FOR: browse_commencement.mov ----------------------------------------- Demonstrates similarity search for photographs of commencement. Starts by searching for a labeled commencement job folder, then picking a seed photograph, and then searching for visually similar photographs. Note that while many of the photographs are of other commencements with crowds of people in robes, later results in the similarity search also returns several photographs of normally-dressed crowds. ----------------------------------------- DATA DESCRIPTION FOR: browse_computer.mov ----------------------------------------- Demonstrates similarity search for photographs of people working on computers. Note that similar results turn up items from across the collection, mostly people working at computers, although some show people working at typewriters, or sitting at a desk with a computer screen in the background. ----------------------------------------- DATA DESCRIPTION FOR: close_match_sets.mov ----------------------------------------- Demonstrates the close match set editor interface. 0:00-0:35 - 3 sets with many very close matches: a portrait, a dual portrait, and an architectural study 0:35-0:38: Pictures of many large crowds erroneously put together as close matches. This is an example where an editor rejects the match and excludes the photographs from future matches. 0:39-048: Close match set that includes images with close matches or duplicates that are horizontally-reversed, but still successfully matched. 0:52-1:04 : A series of pictures from buggy races. Demonstrates opening up photographs to see details in order to determine whether to mark a set of pictures as being true close matches. 1:05-1:30: Another example of buggy photos showing the editor zooming in to the see the photo in detail 1:31-1:40: A long series of a professor in front of a projected screen, all close matches 1:41-2:09: Set of marching band in formation, where the editor selects all the pictures of the band in “CIT” formation but must exclude pictures of slightly different formations that the computer had included for the editor’s consideration 2:10-2:54: Series of commencement photos focused on a speaker at a podium. Some of the photos are of a different speaker and need to be excluded by the editor. ----------------------------------------- DATA DESCRIPTION FOR: tagging_commencement.mov ----------------------------------------- Demonstrates the visual-similarity-enhanced tagging interface 0:00-0:02 - Editor checks out tag 0:03-0:15 - Editor picks a seed photograph 0:16-0:23 - Visually-similar results show up, and editor picks one to go to its associated job folder 0:24-1:06 - Editor assess the entire job, zooming in to detail views on some pictures to distinguish pictures of students processing from those of faculty processing 1:07-1:11 - Editor returns to the visual similarity results, which have now been updated after a bunch of photos were tagged in the previous job. Loads up a new commencement job to tag 1:12-1:39 - Tags the job, whos the use of the “exclude remaining” button to mass reject photos for tagging 1:40-2:02 - Another pic, another job 2:03-2:42 - Another pic, another job 2:43-3:43 - An example of a multi-page job 2:54-2:55 - Showing how close-match photos have their tags linked to get accepted/rejected as a group ----------------------------------------- DATA DESCRIPTION FOR: face_detection.mov ----------------------------------------- Demonstrates the Google Cloud Vision API face detection results browse interface 0:00-0:08 - Paging through results of faces sorted by face detection confidence 0:09-0:28 - Clicking in to a detail view to see the annotations made by google onto the photograph 0:28-0:34 - Showing a clearly-visible, yet untagged face in the same photograph, because Google will only tag a maximum of 10 faces in your images 0:36-0:43 - Faces google has scored as “joy” 0:44-0:47 - Faces google has scored as “sorrow” 0:48-1:07 - Faces google has detected “headwear” in, finishing with a detail shot of a group portrait in a cleanroom (note that google labeled this “cook”!) object_detection.mov 0:00-0:20 - Showing lots of neckties 0:21-0:25 - mis-identification of a polka-dotted shirt as a necktie 0:26-0:45 - More ties including bowties 0:46-1:10 - televisions/monitors in different contexts 1:11-1:41 - tires, including on a buggy and on a tricycle 1:42-1:50 - umbrellas 1:54-2:03 - trumpets, including a detail shot of a music lesson 2:15-2:28 - handbags, including a detail shot that shows google mis-identifying a bandana as a handbag, perhaps because the photograph was rotated 90º