Can one blurred trace reveal a motion’s identity?

A shape appears for thirty moments, partly hidden and stripped of context. Its appearance seems obvious. Its movement does not. Make a first judgment before the other evidence is revealed.

Your first judgment

The class label remains hidden.

Synchronized Oblique signal · partially occluded intensity Frame 01 / 30

Preparing feature traces…

Observation: shape evidence survives; trajectory evidence is incomplete.

18 × 18 · feature trace

0.00 s

Why this archive exists

A dynamic identity is neither shape alone nor motion alone.

Every observed class shares its shape with one class and its motion with another. A single feature trace therefore leaves a plausible rival. The archive is designed to let complementary evidence resolve that ambiguity over time.

Same shapeCircle × horizontalorCircle × ellipse

Same motionCircle × horizontalorCross × horizontal

Eight identities

Shape × motion field

Only eight of the sixteen possible combinations are observed. Choose one to replay its five synchronized traces.

Fifty appearances

A class is a family, not a copied clip.

Phase, amplitude, speed, scale, pose, contrast, occlusion and feature noise vary independently. Four sequences per class deliberately approach the paired-motion boundary.

46 nominal sequences4 boundary sequences

Open case archive

Inspect it. Reproduce it. Question it.

A controlled dynamic collection containing 400 sequences, eight classes, five synchronized signal families and thirty time points. Every NumPy block carries the same sample order, labels and sequence parameters.

Loading file manifest…
Integrity manifest · README

Sequences: 400
Classes: 8
Per class: 50
Time points: 30

Aligned frames256 features · loading Oblique frames324 features · loading Temporal events512 features · loading Spatial spectrum64 features · loading Motion fields288 features · loading

PYTHON / NUMPY

import numpy as np
B = np.load('aligned_frames.npz', allow_pickle=False)
signal = B['data']  labels = B['class_id']