跳转到内容

Motion Magnitude

Motion Magnitude 量化生成视频包含的内部运动量,与运动是否正确无关。它检测保守生成——视觉稳定但近乎静止的输出——不评估方向、语义或物体行为。

对相邻帧 (It,It+1)(I_t, I_{t+1}),SEA-RAFT 给出光流幅值 Ft(x,y)=ut2+vt2F_t(x,y) = \sqrt{u_t^2 + v_t^2}。帧对统计量为像素中位数;原始得分对全部帧对取均值:

Smagraw=1T−1∑t=0T−2medianx,y Ft(x,y)(1)S_{\mathrm{mag}}^{\mathrm{raw}} = \frac{1}{T-1}\sum_{t=0}^{T-2} \mathrm{median}_{x,y}\, F_t(x,y) \tag{1}

WorldAtlas 以 μ=3.2425\mu = 3.2425、σ=3.4505\sigma = 3.4505 做 z-score 校准,裁剪至 [−0.8638,2.7498][-0.8638, 2.7498],映射至 [0.25,0.75][0.25, 0.75]。归一化得分越高表示可见运动越强。计算成功时为 official。

视频评测对所有模型采用相同的输入规则:保持宽高比,长边最多 1920 像素,小视频不放大。SEA-RAFT 仍使用原有的半尺寸推理,因此网络输入长边最多 960 像素;全部相邻帧对仍参与统计。计算幅值和像素中位数前,按两个轴分别将光流换算回原视频像素单位:u=ueWs/Weu=u_e W_s/W_e、v=veHs/Hev=v_e H_s/H_e。报告记录原尺寸、评测尺寸和两个缩放系数。此换算保留位移单位,但改变推理分辨率不保证与未限尺寸的结果数值相同;受影响的旧分数应重算,不能混入同一个汇总结果。

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

输入图像: image_dynamic_photorealistic_articulated_0004
输入图像
输入已核验 · 同一 benchmark case · gap 0.6749

image_dynamic_photorealistic_articulated_0004

套件: image_dynamic场景: photorealistic / articulated
基准 prompt target

Camera fixed. A white horse gallops through a snowy field, legs extending and gathering as its hooves kick up loose snow. Its mane and tail flow while the snow-covered trees remain still.

prompt source: prompt_target

两条视频的首帧都已通过与展示输入的几何匹配核验;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 0.9149

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 0.2400
输入图像: image_dynamic_photorealistic_articulated_0068
输入图像
输入已核验 · 同一 benchmark case · gap 0.7604

image_dynamic_photorealistic_articulated_0068

套件: image_dynamic场景: photorealistic / articulated
基准 prompt target

Camera fixed. The dog runs quickly across the grassy field, moving forward with alternating legs, open mouth, and perked ears while the fence and trees stay in place.

prompt source: prompt_target

两条视频的首帧都已通过与展示输入的几何匹配核验;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 1.0000

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 0.2396
输入图像: image_dynamic_photorealistic_articulated_0023
输入图像
输入已核验 · 同一 benchmark case · gap 0.5764

image_dynamic_photorealistic_articulated_0023

套件: image_dynamic场景: photorealistic / articulated
基准 prompt target

Camera fixed. A large herd of horses gallops across the dry landscape, hooves kicking dust while manes and tails stream backward. The two distant observers and low hills remain stable in the background.

prompt source: prompt_target

两条视频的首帧都已通过与展示输入的几何匹配核验;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 0.8167

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 0.2403