Tips and Techniques/Workflow/General Rotoscoping Principles

From VFXPedia

< Tips and Techniques
Revision as of 19:31, 22 June 2011 by Tilt (Talk | contribs)
(diff) ← Older revision | Current revision (diff) | Newer revision → (diff)
Jump to: navigation, search

1. First of all, look carefully at the whole sequence to define which objects overlap and move to different directions. All these objects require separate masks to avoid excessive keyframing and lots of vertices which aren't used in most of frames. The masks should overlap each other so that there are no seams between them.

Generally, a full-lenth walking or running person should get separate masks for each leg, arm, for the body and the head. If the relative position of the head and the body as well as the way they overlap don't change a lot, you can pull a common mask for both. If the arms don't overlap, they can also be masked by a single shape. If the forearm rotates and goes in front of the arm, they should be masked separately.

If you have a close-up of a turning head, it's better to mask the nose and ears separately, otherwise in the frames where they are hidden or appear from the other side of the head, you'll end up with excessive amount of vertices which have to be arranged into a smooth line.

If you rotoscope a car, it's always a good idea to create separate masks for all the projecting details like mirrors and lamps and for the parts you see through the windows.

It's really important to mask objects themselves rather than make a common contour and cut the holes between them. For example, if a person waves a flag holding it by both hands, the shape of the arms, body, head and flagpole will be more constant and easy to mask than shape of the spaces between them.

Don't just blindly roto every 10th frame or every 2nd. Start with large steps between keys but put them where movement changes happen! (e.g where an arm changes direction or starts to swing into the opposite direction). Then create intermediate frames but likewise not just halfway between existing keyframes. Instead, put them where your shape differs the most from the plate.

A useful technique is to stabilize the camera movement beforehand and put the tracking data back onto the masks. You can also try to stabilize the object if possible (e.g the eyes if you roto a head). This way, you will only have to keyframe the actual motion and shape changes of your subject while camera shakes and jittering are taken care of by the trackers.

Move as many spline points at the same time as possible instead of moving them individually. Use the shortcuts ctrl-B, T or S if you have selected multiple vertices.

Consider using BSplines for organic shapes.

If you notice that you don't need to roto a part of a shape because it's in front of another mask for a while (e.g. arm in front of another person's torso), roto anyway - at least roughly. In case the arm moves outside again this will help you get the shape back together and the points keep their relative position to each other.

Turn masks off or reduce the level to zero when finished. Don't just move them offscreen. Motion blur and DoD will thank you.