Skip to content
WorldAtlas WorldAtlas WorldAtlas
Home

Segment Continuity

Segment Continuity measures whether a generated video contains detected scene cuts. Unintended hard transitions break temporal coherence and indicate segment discontinuities.

Its scope is scene_cut_detection_only. A no-cut result does not establish object identity, spatial-layout consistency or absence of motion freezes; use those abilities’ own measurements separately.

TransNetV2 outputs a per-frame scene-boundary probability pip_i for decoded frame xix_i. Frames with pi>θp_i > \theta are flagged as cut candidates, with defaults θ=0.5\theta = 0.5. A cut is recorded at frame index ii when the binary mask rises from 00 to 11 at ii, following TransNetV2 transition parsing.

To suppress spurious detections, cut candidates are filtered so that every retained scene spans at least LL frames (default L=10L = 10). Let ncutn_{\mathrm{cut}} denote the number of retained cuts in one video. The per-video score is binary:

Sseg={1,ncut=00,ncut>0(1)S_{\mathrm{seg}} = \begin{cases} 1, & n_{\mathrm{cut}} = 0 \\ 0, & n_{\mathrm{cut}} > 0 \end{cases} \tag{1}

Across NN evaluated videos, the model-level aggregate is

Sˉseg=1−ncutN(2)\bar{S}_{\mathrm{seg}} = 1 - \frac{n_{\mathrm{cut}}}{N} \tag{2}

where ncutn_{\mathrm{cut}} counts videos with at least one retained cut. WorldAtlas stores the per-sample binary value from Eq. (1) as the metric raw score. The metric is prediction-only and requires TransNetV2 inference code and weights.

Verified Same-Input Examples

Section titled “Verified Same-Input Examples”

Audit manifest: docs/assets/metric_examples/curated_same_case_manifest.json.

Input image: image_static_photorealistic_city_0064
Input image
Input verified · same benchmark case · gap 1.0000

image_static_photorealistic_city_0064

Suite: image_staticScene: photorealistic / outdoor / city
Benchmark prompt target

Camera pull out. An expanded city view with buildings, distant hills, and extended waterfront.

prompt source: prompt_target

Both first frames passed geometric matching against the displayed input; labels compare this metric only.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 1.0000Detected hard cuts: 0

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.0000Detected hard cuts: 1
Input image: image_static_photorealistic_aquatic_landscape_0084
Input image
Input verified · same benchmark case · gap 1.0000

image_static_photorealistic_aquatic_landscape_0084

Suite: image_staticScene: photorealistic / outdoor / aquatic_landscape
Benchmark prompt target

Camera move left. A lake edge with reeds, autumn trees, and distant mountains.

prompt source: prompt_target

Both first frames passed geometric matching against the displayed input; labels compare this metric only.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 1.0000Detected hard cuts: 0

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.0000Detected hard cuts: 1
Input image: image_static_photorealistic_aquatic_landscape_0089
Input image
Input verified · same benchmark case · gap 1.0000

image_static_photorealistic_aquatic_landscape_0089

Suite: image_staticScene: photorealistic / outdoor / aquatic_landscape
Benchmark prompt target

Camera pull out. A rocky shoreline with trees and distant hills. Camera move right. A forest edge with dense foliage and sunlight. Camera pan right. A forest edge with forest path, scattered rocks, and tree canopy.

prompt source: prompt_target

Both first frames passed geometric matching against the displayed input; labels compare this metric only.

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Higher metric scoreScore 1.0000Detected hard cuts: 0

Model result videos will be available with the ModelScope release. / 模型推理视频将在 ModelScope 方案上线后提供。

Lower metric scoreScore 0.0000Detected hard cuts: 1

References

Section titled “References”