时序闪烁（Temporal Flickering）

时序闪烁度量生成视频相邻帧之间的像素稳定性。帧间高频亮度或颜色跳变会产生可见闪烁；本指标采用 RGB 像素平均绝对误差（MAE）加以惩罚。

设解码帧序列为 $\{x_i\}_{i=1}^{N}$，相邻帧像素 MAE 为

$$\mathrm{MAE}(x_i, x_{i+1}) = \frac{1}{HWC}\sum_{h,w,c}\left|x_i(h,w,c)-x_{i+1}(h,w,c)\right| \tag{1}$$

其中 $H$、$W$、$C$ 分别为高、宽与通道数。WorldAtlas 在 uint8 RGB 取值区间 $[0,255]$ 上计算 MAE；若分辨率不一致，则将后一帧缩放至与前一帧相同后再作差。

时序闪烁得分对相邻帧 MAE 取平均：

$$S_{\mathrm{flick}} = \frac{255 - \frac{1}{N-1}\sum_{i=1}^{N-1}\mathrm{MAE}(x_i, x_{i+1})}{255} \tag{2}$$

$S_{\mathrm{flick}}$ 越高表示闪烁越少。该指标仅需 prediction，且至少需要两帧。

已核验的同输入案例

Audit manifest: docs/assets/metric_examples/missing_pages_examples_20260824.json.

输入图像

输入已核验 · 同一 benchmark case · gap 0.3521

image_static_stylized_passageways_0134

套件: image_static场景: stylized / indoor / passageways

基准 prompt target

Camera pan right. An office hallway with office doors, bulletin board, and people. Camera move right. An office lounge area with couches, coffee table, and plants. Camera pull out. An office lounge area with entrance, reception desk, and waiting area.

prompt source: prompt_target

两条视频属于同一 benchmark sample；高低标签只比较当前 metric。

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。

较高指标分分数 0.9907

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。

较低指标分分数 0.6386

输入图像

输入已核验 · 同一 benchmark case · gap 0.3323

image_static_stylized_city_0123

套件: image_static场景: stylized / outdoor / city

基准 prompt target

Camera pull out. A park path with benches, trees, and wider path. Camera move right. A city street with buildings, parked cars, and benches. Camera pan right. An urban square with fountain, people, and shops.

prompt source: prompt_target

两条视频属于同一 benchmark sample；高低标签只比较当前 metric。

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。

较高指标分分数 0.9766

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。

较低指标分分数 0.6443

输入图像

输入已核验 · 同一 benchmark case · gap 0.2363

image_static_photorealistic_verdant_landscape_0028

套件: image_static场景: photorealistic / outdoor / verdant_landscape

基准 prompt target

Camera pan left. A building facade with trees, pine needles, and car. Camera pull out. A street view with parked cars, pedestrians, and building facade. Camera move left. A street view with city park, benches, and people.

prompt source: prompt_target

两条视频属于同一 benchmark sample；高低标签只比较当前 metric。

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。

较高指标分分数 0.9873

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。

较低指标分分数 0.7510

参考文献

VBench — Temporal Flickering（附录 C.2.3）

相邻帧像素 MAE 经 $(255 - \overline{\mathrm{MAE}})/255$ 映射至 $[0,1]$ 稳定性得分。