首页 > AI前沿 > Training-free motion statistics match self-supervised behavior encoders

Training-free motion statistics match self-supervised behavior encoders

arXiv机器学习 2026-11-19 13:00 16 阅读 查看原文
用户要求我作为专业的新闻排版工程师,将提供的纯文本新闻正文排版成语义化的HTML片段。 规则: 1. 只输出HTML片段本身,不要Markdown围栏、不要解释。 2. 每个自然段落用

包裹,长文用

/

划分小节。 3. 列表内容用
  • ,引用用
    ,代码用
    ,重要句子可用。
    4. 保留原文所有信息,不得增删改内容。
    5. 不要输出//等文档级标签。
    
    正文内容分析:
    这是一段关于动物行为自监督编码器的科学研究新闻/摘要。
    原文只有一个长段落。我需要将其合理划分为小节,并用

    或

    来划分,同时保留所有信息。 原文拆分思考: "Self-supervised encoders for animal behavior are often trained on, or fed, hand-built movement features. We ask whether they beat those ingredients when compared fairly." -> 引言/背景部分 "A training-free representation, made of generic per-frame measurements summarized over fixed time windows and reduced with PCA, is compared with retrained and released encoders (BAMS, hBehaveMAE) under a preregistered protocol: frozen representations of equal size and one shared classifier." -> 方法/实验设置 "On CalMS21 it reaches 85.8 macro F1 against 82.4 for the best encoder, ties with four labelled videos, and leads when the annotator changes or new behaviors must be learned from a few videos; the reading holds for encoders retrained with three seeds." -> CalMS21结果 "On the MABe22 mouse triplets it scores 83.7 against 80.3 for the released hBehaveMAE (gap +3.4, 95% interval +2.1 to +4.7)." -> MABe22 mouse结果 "On a second species, the MABe22 fly groups, the same recipe scores 51.6 mean F1 over 50 tasks against 48.2 for BAMS, the best published method." -> MABe22 fly结果 "On Shot7M2, the synthetic basketball benchmark built by hBehaveMAE's authors, it scores 68.8 against 64.7 for hBehaveMAE, the best method in that paper, though hBehaveMAE remains better on the shortest movements." -> Shot7M2结果 "Short time windows carry most of the result." -> 补充说明 "Raw keypoints in each body's own frame, summarized over the same windows, already match or beat the encoders on both mouse datasets and on Shot7M2; on the fly groups they fall short, and the measurements add 5.3 points." -> 原始关键点分析 "On these benchmarks, current behavior encoders have yet to beat the hand-built features they learn from, and such baselines should be reported alongside them." -> 结论 排版设计: 可以加一个总标题

    ,然后分几个小节

    。 或者直接把原文分成几个段落,加上小标题。

    研究背景与实验设计

    Self-supervised encoders for animal behavior are often trained on, or fed, hand-built movement features. We ask whether they beat those ingredients when compared fairly. A training-free representation, made of generic per-frame measurements summarized over fixed time windows and reduced with PCA, is compared with retrained and released encoders (BAMS, hBehaveMAE) under a preregistered protocol: frozen representations of equal size and one shared classifier.

    基准测试结果

    On CalMS21 it reaches 85.8 macro F1 against 82.4 for the best encoder, ties with four labelled videos, and leads when the annotator changes or new behaviors must be learned from a few videos; the reading holds for encoders retrained with three seeds. On the MABe22 mouse triplets it scores 83.7 against 80.3 for the released hBehaveMAE (gap +3.4, 95% interval +2.1 to +4.7). On a second species, the MABe22 fly groups, the same recipe scores 51.6 mean F1 over 50 tasks against 48.2 for BAMS, the best published method. On Shot7M2, the synthetic basketball benchmark built by hBehaveMAE's authors, it scores 68.8 against 64.7 for hBehaveMAE, the best method in that paper, though hBehaveMAE remains better on the shortest movements.

    特征分析与结论

    Short time windows carry most of the result. Raw keypoints in each body's own frame, summarized over the same windows, already match or beat the encoders on both mouse datasets and on Shot7M2; on the fly groups they fall short, and the measurements add 5.3 points. On these benchmarks, current behavior encoders have yet to beat the hand-built features they learn from, and such baselines should be reported alongside them.

    检查规则: 1. 只输出HTML片段,无Markdown围栏,无解释。 -> 是 2. 每个自然段落用

    包裹,长文用

    /

    划分小节。 -> 是,用了

    和

    3. 列表、引用、代码、重要句子。 -> 原文没有列表、引用、代码,重要句子用了。 4. 保留原文所有信息,不得增删改内容。 -> 检查文本是否完全一致。 原文:"Self-supervised encoders..." 我的版本:

    Self-supervised encoders for animal behavior are often trained on, or fed, hand-built movement features. We ask whether they beat those ingredients when compared fairly. A training-free representation, made of generic per-frame measurements summarized over fixed time windows and reduced with PCA, is compared with retrained and released encoders (BAMS, hBehaveMAE) under a preregistered protocol: frozen representations of equal size and one shared classifier.

    完全一致。

    On CalMS21 it reaches 85.8 macro F1 against 82.4 for the best encoder, ties with four labelled videos, and leads when the annotator changes or new behaviors must be learned from a few videos; the reading holds for encoders retrained with three seeds. On the MABe22 mouse triplets it scores 83.7 against 80.3 for the released hBehaveMAE (gap +3.4, 95% interval +2.1 to +4.7). On a second species, the MABe22 fly groups, the same recipe scores 51.6 mean F1 over 50 tasks against 48.2 for BAMS, the best published method. On Shot7M2, the synthetic basketball benchmark built by hBehaveMAE's authors, it scores 68.8 against 64.7 for hBehaveMAE, the best method in that paper, though hBehaveMAE remains better on the shortest movements.

    完全一致。

    Short time windows carry most of the result. Raw keypoints in each body's own frame, summarized over the same windows, already match or beat the encoders on both mouse datasets and on Shot7M2; on the fly groups they fall short, and the measurements add 5.3 points. On these benchmarks, current behavior encoders have yet to beat the hand-built features they learn from, and such baselines should be reported alongside them.

    完全一致。注意,加上标签没有改变文字内容。 5. 不要输出//等文档级标签。 -> 是。 微调标题,使其更符合新闻排版:

    研究背景与实验设计

    多基准测试表现

    特征分析与研究结论

    确保没有多余的空行或Markdown。直接输出HTML。 检查是否有遗漏的字符或标点。 "Self-supervised encoders for animal behavior are often trained on, or fed, hand-built movement features. We ask whether they beat those ingredients when compared fairly. A training-free representation, made of generic per-frame measurements summarized over fixed time windows and reduced with PCA, is compared with retrained and released encoders (BAMS, hBehaveMAE) under a preregistered protocol: frozen