本仓库现已清理视觉识别意图相关模块，当前仅保留 IMU 多数据集意态感知流程（JIGSAWS、Opportunity、NinaPro、PAMAP2）。
config/
imu_multidataset.toml
imu_intent/
loaders.py
train_multidataset.py
infer_from_csv.py
stream_demo.py
verify_jigsaws_layout.py
fetch_jigsaws_official.py
eval/
plot_imu_dashboard.py
plot_imu_timeline.py
plot_intent_transition_graph.py
build_visual_gallery.py
docs/
IMU意态感知程序说明.md
IMU可视化展示方案.md
JIGSAWS官方下载接入说明.md
python3 -m pip install -r requirements.txtpython3 -m imu_intent.train_multidataset \
--config config/imu_multidataset.toml \
--model-out models/imu_intent_multidataset_real.joblib \
--metrics-out logs/imu_intent_metrics_real.jsonpython3 -m imu_intent.infer_from_csv \
--model models/imu_intent_multidataset_real.joblib \
--input data/public_imu/PAMAP2_Dataset/Protocol/subject101_small.csv \
--output logs/imu_intent_subject101_small_predictions.csv \
--summary logs/imu_intent_subject101_small_summary.jsonpython3 -m eval.plot_imu_dashboard --metrics logs/imu_intent_metrics_real.json --output plots/imu_dashboard.png
python3 -m eval.plot_imu_timeline --predictions logs/imu_intent_subject101_small_predictions.csv --output plots/imu_timeline.png
python3 -m eval.plot_intent_transition_graph --predictions logs/imu_intent_subject101_small_predictions.csv --output plots/imu_transition_graph.png
python3 -m eval.build_visual_gallery --plots-dir plots --output plots/gallery.html增强版说明：
- 使用 151 个可解释运动学特征（速度/加速度/jerk 分布、双手协同、时间形态与能量熵）
- 引入手术任务上下文（
Knot_Tying/Needle_Passing/Suturing） - 自动比较
RandomForest、ExtraTrees、XGBoost，并做分组交叉验证下的取交集 - 在不看测试集的前提下，仅用训练集做术者分组 OOF 选择，再用全数据重拟合部署模型
- 当前严格 OOF 结果：Accuracy
0.8417，Macro-F10.8266，ROC-AUC0.8958 - 当前自动选中方案：
XGBoost + ExtraTrees贪心集成，叠加因果EMA(alpha=0.75)时序平滑
python3 -m imu_intent.jigsaws_intent_program \
--config config/jigsaws_intent.toml \
--output-dir logs/jigsaws_intent \
--model-out models/jigsaws_intent_model.joblib
python3 -m eval.plot_jigsaws_intent_report \
--metrics logs/jigsaws_intent/metrics.json \
--predictions logs/jigsaws_intent/window_predictions.csv \
--output plots/jigsaws_intent_report.png
python3 -m eval.plot_jigsaws_showcase \
--metrics logs/jigsaws_intent/metrics.json \
--predictions logs/jigsaws_intent/window_predictions.csv \
--output-dir plots/jigsaws_showcase
python3 -m eval.plot_jigsaws_roc \
--predictions logs/jigsaws_intent/window_predictions.csv \
--metrics logs/jigsaws_intent/metrics.json \
--output-dir plots/jigsaws_roc详细说明见：
docs/JIGSAWS手术意态感知程序.md
官方入口：
收到官方邮件下载链接后：
python3 -m imu_intent.fetch_jigsaws_official \
--suturing-url "<官方邮件链接1>" \
--knot-url "<官方邮件链接2>" \
--needle-url "<官方邮件链接3>"
python3 -m imu_intent.verify_jigsaws_layout \
--config config/imu_multidataset.toml \
--output logs/jigsaws_layout_check.json