REQUESTED VS. DELIVERED LIVE EXPERIMENT
The pharmacist
and the generator.
The pharmacist turns each request into a scene a renderer can execute. The generator renders the image pair with correspondence labels.
Sketch the motion
Draw your own arrows, or load a preset below. Presets are starting points you can edit.
A sketch is a request, not a guarantee. Compare the motion you asked for with what the scene delivers.
Meet your scene
The starting view
The destination view
Requested vs. delivered motion
A → B · SAME PIXEL SCALEInput arrows sit on top of a sparse sample of valid foreground flow. Both use their original positions and lengths.
Generate a pair to compare its labels with your sketch.
Two images.
Pixel-level labels.
Your scene recipe.
Ground each request
in a realizable scene.
The arrows describe positions and displacements. A trained inverse model—the pharmacist—maps this set to camera and object parameters. The renderer makes the pair and computes correspondences from the scene’s surfaces.
These are synthetic scenes from the research pipeline. Labels describe the actual rendered motion, including visibility; they do not copy your sketch. Arbitrary drawings may fall outside what the model can realize.