This page last changed on Jul 23, 2009 by dcline.

Attendees: D.Cline, M.Vadaro, K.Smith, L.Kuhnz

Individual updates:

Ken and Mike are on travel starting ~8/2/09 to Angola. Ken will be back 8/14/09 and Mike will be gone all of August.

First 100 ft section of a roll was digitized (thanks Jake !) which took around 11 hours and resulted in 671 images. Three additional 100 ft sections will be the roll will be digitized in the coming weeks and used to extract a training data set.

Linda presented some Photoshop batch processing to erase the timecode overlay and crop black bars out on the first images from the 100 ft roll. There is some variation in both width and height of the black bars on each frame, making the batch processing difficult, but she did a great job, and reports only ~8% loss of pixels over the masked timecode area from the processing.

Danelle finished installing AVED software on several machines, and is working on some modifications to the AVED scripts to handle still frames, instead of video. She will send some information to Linda on frame sizes and resolutions to test and an ISO standard naming convention to use.

The raw digitized images are large at 4504x2790 and 72 pixels per inch. AVED has never processed images these large. There may be limitations on computational time and/or memory so we will be doing some test on these to see if we need to scale them back to HD size, which may be more reasonable. All previous testing was done on resized 1120x720 images.

More pixels are generally better for AVED classification. There are no hard cut lines, because it depends on the animal, its features etc.. However, based on previous classification testing on images of Poebious, it was found that images smaller than 20x20 pixels are be too small for effective classification. Linda will find an animal that is reasonable to count and get an average pixel size to determine how much we could scale down images, if needed, before loosing accuracy.

Lots of interesting discussion in this meeting on how the film was analyzed, the importance of grid overlays, whether the black images (strobe mis-fires) are included in the digitized reels, etc.

Document generated by Confluence on Feb 03, 2026 15:34