SensorQA
收藏资源简介:
SensorQA是由乔治亚理工学院和加州大学圣地亚哥分校的研究团队创建的首个用于日常监测的问答数据集。该数据集包含5648个多样且实用的问答对,反映了真实的人类兴趣,并与传感器数据中的准确答案配对。数据集基于60名用户在长达三个月的时间内收集的传感器数据,涵盖了从一天到多周的不同时间尺度。数据集的创建过程包括通过亚马逊机械土耳其平台(AMT)生成问答对,确保问题的多样性和实用性。SensorQA的应用领域主要集中在健康监测和日常活动分析,旨在通过问答形式帮助用户从传感器数据中提取有用的信息。
SensorQA is the first question answering dataset for daily monitoring created by research teams from Georgia Institute of Technology and University of California, San Diego. It contains 5,648 diverse and practical question-answer pairs that reflect real human interests and are paired with accurate answers sourced from sensor data. The dataset is based on sensor data collected by 60 users over a three-month period, covering various time scales ranging from one day to multiple weeks. The dataset creation process included generating question-answer pairs via the Amazon Mechanical Turk (AMT) platform to ensure the diversity and practicality of the questions. The application fields of SensorQA mainly focus on health monitoring and daily activity analysis, aiming to help users extract useful information from sensor data through the question answering format.
SensorQA 数据集概述
数据集简介
SensorQA 数据集旨在训练模型理解可穿戴设备的传感器读数,并能够回答相关问题。该数据集使用了 Extrasensory 数据集中的传感器读数和特征。
数据集内容
-
标注文件:
SensorQA/overall_sensorqa_dataset_train.jsonSensorQA/overall_sensorqa_dataset_train_em.jsonSensorQA/overall_sensorqa_dataset_val.jsonSensorQA/overall_sensorqa_dataset_val_em.json
-
传感器读数的图形可视化:
SensorQA/non_oracle_graphsSensorQA/oracle_graphs
数据集来源
- 传感器读数和特征来自 Extrasensory 数据集,可通过以下链接获取:http://extrasensory.ucsd.edu/
基线模型性能
对话数据集基线
| 模态 | 骨干模型 | ZS/FT | Oracle | Rouge-1 | Rouge-2 | Rouge-L | Meteor | Bleu | Exact Match |
|---|---|---|---|---|---|---|---|---|---|
| L | T5-Base | FT | 0.71 | 0.55 | 0.69 | 0.70 | 0.43 | 0.26 | |
| L | Llama-7B-LORA | FT | 0.72 | 0.62 | 0.72 | 0.72 | 0.38 | 0.04 | |
| V+L | Llama-7B-Adapter | ZS | ✔ | 0.33 | 0.20 | 0.30 | 0.44 | 0.09 | 0 |
| V+L | Llama-7B-Adapter | FT | ✔ | 0.73 | 0.57 | 0.71 | 0.72 | 0.43 | 0.14 |
| V+L | Llava-1.5-LORA | FT | ✔ | 0.62 | 0.46 | 0.60 | 0.58 | 0.35 | 0.13 |
| V+L | Llama-7B-Adapter | ZS | ✘ | 0.09 | 0.42 | 0.31 | 0.19 | 0.28 | 0 |
| V+L | Llama-7B-Adapter | FT | ✘ | 0.43 | 0.72 | 0.73 | 0.57 | 0.70 | 0.14 |
| V+L | Llava-1.5-LORA | FT | ✘ | 0.64 | 0.47 | 0.61 | 0.60 | 0.35 | 0.11 |
| S+L | Llama-7B-Adapter-HC | FT | 0.72 | 0.55 | 0.70 | 0.71 | 0.42 | 0.14 | |
| S+L | Llama-7B-Adapter-CLIP | FT | 0.71 | 0.53 | 0.69 | 0.69 | 0.40 | 0.12 |
精确匹配基线
| 模态 | 骨干模型 | ZS/FT | Oracle | 准确率 |
|---|---|---|---|---|
| L | T5-Base | FT | 25.4% | |
| L | Llama-7B-LORA | FT | 26.5% | |
| V+L | Llama-7B-Adapter | ZS | ✔ | 0% |
| V+L | Llama-7B-Adapter | FT | ✔ | 28% |
| V+L | Llava-1.5-LORA | FT | ✔ | 21.5% |
| V+L | Llama-7B-Adapter | ZS | ✘ | 0% |
| V+L | Llama-7B-Adapter | FT | ✘ | 26.2% |
| V+L | Llava-1.5-LORA | FT | ✘ | 11% |
| S+L | Llama-7B-Adapter-CLIP | FT | 23.5% | |
| S+L | Llama-7B-Adapter | FT | 24.8% | |
| S+L | DeepSQA | FT | 27.46% |
复现基线模型的步骤
-
T5 基线: bash python question_only.py python t5_text_evaluation
-
LLama-7B-LORA 实验: bash python llama_lora_training.py --train python llama_lora_training.py --eval python llama_text_evaluation.py
-
Llama-7B-Adapter 视觉+语言微调:
- 调整配置文件
Llama-Adapter/llama_adapter_v2_multimodal7b/exps/finetune-data-config.yaml - 需要 llama 权重和 adapter 权重
- 参考链接:https://github.com/OpenGVLab/LLaMA-Adapter/blob/main/llama_adapter_v2_multimodal7b/docs/train.md
- 运行命令: bash cd LLaMA-Adapter/llama_adapter_v2_multimodal7b ./exps/finetune.sh models/llama LLaMA-Adapter/ckpts/7fa55208379faf2dd862565284101b0e4a2a72114d6490a95e432cf9d9b6c813_BIAS-7B.pth exps/finetune-data-config.yaml outputs python llama_adapter_val_loop.py python llama_text_evaluation.py
- 调整配置文件
-
Llava-1.5-LORA 结果: bash accelerate launch --mixed_precision fp16 llama_lora_train.py --dataset_name="HuggingFaceH4/llava-instruct-mix-vsft" --model_name_or_path="llava-hf/llava-1.5-7b-hf" --report_to="none" --learning_rate=2e-5 --per_device_train_batch_size=1 --gradient_accumulation_steps=1 --output_dir="data/vsft-llava-1.5-7b-hf" --num_train_epochs=4 --gradient_checkpointing --remove_unused_columns=False --torch_dtype=float16 --fp16=True --use_peft=True --lora_r=64 --lora_alpha=16 --lora_target_modules=all-linear --log_level="info" --logging_strategy="steps" --logging_steps=1 python sensorqa_llava_eval.py
-
Llama-7B-Adapter S+L 手工特征结果:
-
下载手工特征:http://extrasensory.ucsd.edu/data/primary_data_files/ExtraSensory.per_uuid_features_labels.zip
-
安装自定义 timm 版本: bash cd pytorch-image-models pip install -e .
-
训练命令: bash cd LLaMA-Adapter/llama_adapter_v2_multimodal7b_sensors ./exps/finetune.sh models/llama LLaMA-Adapter/ckpts/7fa55208379faf2dd862565284101b0e4a2a72114d6490a95e432cf9d9b6c813_BIAS-7B.pth exps/finetune-data-config.yaml outputs python llama_adapter_sensors_val_loop.py python llama_text_evaluation.py
-
-
Llama-7B-Adapter S+L CLIP 特征:
-
下载原始时间序列数据:http://extrasensory.ucsd.edu/data/raw_measurements/ExtraSensory.raw_measurements.raw_acc.zip
-
训练命令: bash cd clip_training ./run.sh
-
训练完成后: bash cd LLaMA-Adapter/llama_adapter_v2_multimodal7b_sensors_clip ./exps/finetune.sh models/llama LLaMA-Adapter/ckpts/7fa55208379faf2dd862565284101b0e4a2a72114d6490a95e432cf9d9b6c813_BIAS-7B.pth exps/finetune-data-config.yaml outputs python llama_adapter_sensors_val_loop.py python llama_text_evaluation.py
-
-
DeepSQA 模型:
- 参考代码库:https://github.com/nesl/DeepSQA
- 训练命令: bash cd DeepSQA python3 deepsqa_ca.py --gpt_shortened

- 1SensorQA: A Question Answering Benchmark for Daily-Life Monitoring乔治亚理工学院, 加州大学圣地亚哥分校 · 2025年



