FORM: Best Annotation Practices


Techniques for viewing and controlling the video

* Always look at gestures both frame-by-frame and also in real-time.
  It is possible to confuse one movement for a similar movement if it is only 
  viewed at one speed.  For example, a "wrist circular" hand movement looks very 
  similar to a "down-up" movement combined with a "toward thumb-toward little   
  finger" movement when only viewed frame-by-frame.  However, the circular motion 
  is apparent when the clip is played at normal speed.
* If helpful, look at the video in full screen; motion can be lost when looking 
  at a small version of the video.


Techniques for looking at the speaker

* Use reference points, shadows and blurs to help determine where movement starts,
  stops and changes direction.
* Often it can be difficult to determine whether a movement results from actual 
  arm movement, or simply a change in body position.  If a frame of video is blurred 
  around the hand or arm, it signifies that there is indeed movement.  Likewise, 
  inferences about movement can be made from the relative position of a shadow to the 
  body or the background.  This can help in determining not only if there is movement, 
  but also a change in the direction of movement.
* The point where the hand or arm begins to blur is often the point where it starts to 
  move.
* Changes in the wrinkles of clothing can signify movement.
* Much like blurs and reference points the clothing of the subject can give indications 
  to aid in determining the movement that is occurring.
* If the subject's clothing is moving around the elbow, then it signifies elbow movement.
  Likewise, changes in the size and/or number of creases or folds around the shoulder 
  would signify upper arm movement.  This is most helpful in distinguishing which part 
  of the arm is exerting the effort in a gesture.
* Be careful to look at body parts separately.  
* While all parts of the arm work together, it is easy to lose upper arm motion and 
  instead attach the movement to the forearm, or lose forearm movement to wrist movement,
  for example.  One way to account for this is to cover the part of the arm that you are 
  not looking at.
* Similarly, be careful to distinguish between wrist movement and small forearm rotations. 
  They often occur simultaneously, but it can be difficult to distinguish when the rotation 
  is small.

How to use the FORM scheme

* When choosing a hand shape, it is often difficult to find an exact match.  In this case, 
  tension may compensate for the difference.  For example, a loose fist could be annotated 
  as a relaxed version of a tight fist if it is the most similar shape.
* Compare how far your wrist bends towards your thumb with how far it bends towards your 
  little finger.  Use extremely towards little finger in cases where the wrist is bent 
  further towards the little finger than it could bend towards the thumb.
* In the two-handed gestures track, only one of the Boolean values (interlaced, moving in 
  parallel, moving apart etc.) can be chosen per object.
* Hand position is annotated as a location track.  
* When hand position is not changing, a multi-frame object should be created to indicate 
  position.  
* Remember that it is possible for other parts of the arm to be moving without hand position 
  changing.
* Single frame objects are used to indicate changes in direction.  
* This includes the peaks of curved movements.  
* It is sometimes necessary to use more than one object at different points within one 
  continuous motion in order to properly capture the path of the motion.
* Often the hand will move within one box, so consecutive objects will have identical 
  attributes.  Annotate this as motion rather than unchanging location.  If you wish to 
  divide the body using different increments at a later time, the necessary objects will 
  already exist at the proper timestamps.


Annotation Techniques

* Use your own body for reference.
* If the movement is ambiguous due to video quality or body position, it may be helpful 
  to mimic the subject's movement.  Performing the movement can help determine the precise 
  action that is taking place and give an additional perspective to aid in annotation.  
  When you cannot see a movement clearly, always mark it obscured to indicate your uncertainty.  
* The wrist bone's connected to the arm bone...
* The best point of reference for a particular part of the arm is the joint connected to it.  
  When annotating upper arm movement, looking at the elbow's position will show the upper arm 
  lift.  Likewise, to determine forearm orientation, it is most helpful to look at the direction 
  the palm and wrist are facing.  
* Use multiple strokes whenever possible.
* If part of a gesture is repeated in exactly the same manner, annotate it as one movement track 
  and use the "strokes" value to indicate the number of times the action is repeated.
* If you have trouble deciding on values involving angles (upper arm lift, elbow flexion), make 
  a "cheat sheet" with examples of several different angles (e.g. 45, 90, 135, and maybe a few 
  in between), and keep it handy while you annotate.  You can do this easily with a pen and a 
  plastic protractor.
* When selecting a value for effort, it is often helpful to consider the amount of movement 
  compared to the amount of time.  
* A small movement that spans an entire second requires much less effort than a large one that 
  spans a few frames.  
* Viewing the video in real-time may help to make this decision.
* It can be difficult to visualize the dividing line between two segments.  In this case, a small 
  degree of relativism may be appropriate when making the judgment.  For example, if the hand is in 
  the "Right" segment of the X dimension, and it moves more to the right, but you are not sure if it 
  is still "Right" or "Far Right," using the latter option is more likely to capture the motion 
  properly.