Shape of Motion and the Perception of Human Gaits

Boyd and Little present a design for an experiment to test the validity of the shape of motion approach to computer gait recoginition by inverting the idea to generate motion fields and having humans decide how well the fields demonstrate walking. A good paper discussing experimental design, it takes the idea from conception through the first pilot study.

Reference Link

@article{Boyd1998
   author = {Jeffrey E. Boyd and James J. Little},
   title = {Shape of Motion and the Perception of Human Gaits},
   book = {Empirical Evaluation Techniques in Computer Vision},
   year = {1998},
   pages = {155--171}
}

- posted by David @ 3:07 PM

A WWW-Accessible Database for 3D Vision Research

Flynn's database, moved from its old location at WSU to OSU, presents a usable set of ground-truthed range image data, some rendered and some images. The ground truth, in this case, is a description of the 3D model. The paper is short and to the point, and, interestingly, references a couple of places where the data has been used: in a paper by Dickinson and Metaxas and another by Sun and Sherrah.

Reference Link

@article{Flynn1998
   author = {Patrick J. Flynn and Richard J. Campbell},
   title = {A WWW-Accessible Database for 3D Vision Research},
   book = {Empirical Evaluation Techniques in Computer Vision},
   year = {1998},
   pages = {148--154}
}

- posted by David @ 2:47 PM

gt4.a4

I thought it was time for another release of the video annotation tool, so, if you would like access to some of the new features, go over and give it a try. As a plus, there is now an OS X application bundle for those of you with OS X 10.2+ and JDK 1.4.1+. Let me know if you test it under 1.4.2, which I don't have access to, yet. With a couple of exceptions (file dialogs, no sheets, one main window) it should act like any other os x application. If you want, I can split it into three windows instead of using the single main pane.

- posted by David @ 3:55 PM

JNI v. J#

So, as Tony impements a scrubbably frame-accurate version of virtualdub, the issue arises as to how best integrate his code into out java codebase for the annotation tool. I'm certainly a proponent of the java approach for now, as I've just spent the last year writing GTv4 in it. I would like to hold off on switching until version 5, or there is a good runtime and set of UI libraries that are available on more platforms. Unfortunately, this may not be possible if the JNI solution proves too slow or too hard to implement, as MPEG-2 decoding, with playback of the audio stream, is required functionality. I think it should be possible, if a hassle, to get the data with a modicum of copying using nio and some careful planning. If not, slow isn't a bad thing, and I could just fall back on copying image data row-by-row.

- posted by David @ 5:50 PM

More Fixes

As Charles continues to improve polygon editing, I've added import of gtf. There are actually two components - import schema and import data. The schema import will also import xml-format schema. It tries to copy in all the schema, but will ignore attributes with the same name as existing attributes. If people ask, I will allow the user to select which attributes to replace, or possibly support renaming. I've also got dragging timeline endpoints, although using it makes me want to drag the lines themselves; I'll probably make dragging inside a line insert a break.

- posted by David @ 6:32 PM

Single-Frame Editing

It looks like it will be a hassle to add a single-frame editing mode. Basically, it will change the meaning of all the existing mutli-frame editing tools, or simply disable them, which makes editing a hassle. Duplicate should duplicate invalid descriptors to the current frame, propagate and interpolate should work, but instead create a new copy of the descriptor in each frame, that sort of thing. So, let's review the current set up for single frame editing:

1. Create a .info file using the following shell script, invoked beneath the directory of .ppms to make into a .info file (can use any static image format. Really, can use any media format, but static images make more sense.)

   #!/bin/sh
   cat >> $x.info <<EOF
   #VIPER_VERSION_3.0
   1
   EOF
   cd $x
   ls *.ppm >> ../$x.info
   mv ../$x.info ./
   cd ..
   

2. Edit the metadata with viper-gt as a normal file.
3. Use convert to split it into a bunch of single-frame sources. This will involve some creative scripting to work: you have to convert it to .gtf and then back to .xml, possibly adding a "SOURCEFILES" line to the .gtf file, if convert doesn't add it automatically.

- posted by David @ 12:32 AM

Update on V4 Bugs

I've managed to get export working, but not import, yet. I've also got the markers working nicer. I tried the stupid way of doing asynchronous decoding, and it looks as though I'll have to do it the harder way to get any improvement in the interface. I still have to fix some bugs with undo/redo, and implement the single-frame-edit mode. I'll try to do the single-frame edit tonight. Then I'm going to work on dragging in the timeline.

Charles has done some work on fixing the box selection/editing problems. Next comes polygon editing, with the resize frame and everything else. I suggested a UI improvement for shift-click deleting points on the polygon - displaying the results of the click in a different color/line style before the user clicks/while the mouse is hovering over the point. I think he'll hold off on ellipses for a while, possibly putting in a quick obox-derived placeholder, to focus on the polygon improvements.

- posted by David @ 6:14 PM

Bugs For Version 4

Charles and I met with Dave to discuss the state of ViPER today. We are trying to get version four of the ground truth tool together for next week. There were a bunch of bugs for both Charles and I to resolve.

My Bugs

• Fixing Schema editor's support for undo/redo when dealing with live data (schema refactoring). 837466
• Import/Export to .gtf files. 887952
• Single-frame-edit mode. 890051
• Marker Improvements: 819672
  • Go To... button on right-click popup, and possibly on a hover-appearing icon or on the footer.
  • Add an Interpolate to Marker -- option.
  • Add labels to user markers (circle with a letter inside?) and improve coloring/display. Ideas: user markers are black, now marker is an arrow.
• Asynchronous decoding of video/ decoding in a separate thread. This should fix problems when dragging the now marker, and might give the interface more pep in general. 889937
• Dragging end points of time lines should be implemented. They should snap to markers, unless shift is held down. 826964
• Should improve right-click menus and tooltips throughout the interface.
• Need to fix the column resizing in the table - sometimes it doesn't resize right, other times it resizes all by itself, and it should always extend to fit in the scroll pane. 889943
• Need to enhance the table's display of selected items. The selected attribute should be a different color, and it should scroll to display the selected attribute when it is picked. 889941

Charles's Bugs

• Polygon dragging is broken. It seems that the polygon-interior test doesn't work for some polygons, like ones with concavities. 889950
• Highlighting needs to be regularly implemented for each data type. This would include making the highlighting highlight only the section to be moved (i.e. not the whole segment on either side of a corner), there should be some highlighting for moving the box (this will be hard for polygons containing self-crossings, if we decide to do 'highlight the inside') and possibly making the highlights somewhat scale invariant.889962
• When selected, the polygons should should have a surrounding frame with handles to resize them, and possibly rotate. 889966
• The polygons should display the points in a different color when selected. There should also be a method for adding new points / deleting existing ones. 889973
• In general, the selection and resize thresholds should be in screen pixels, not in world pixels. 889975
• The deselection should be smarter - right now, I can't even move a point of a polygon that is inside a bounding box. 889958

- posted by David @ 7:16 PM