The Toyota Smarthome Dataset
This page introduces the Toyota Smarthome dataset. Smarthome has been recorded in an apartment equipped with 7 Kinect v1 cameras. It contains the common daily living activities of 18 subjects. The subjects are senior people in the age range 60-80 years old. The dataset has a resolution of 640×480 and offers 3 modalities: RGB + Depth + 3D Skeleton. The 3D skeleton joints were extracted from RGB. For privacy-preserving reasons, the face of the subjects is blurred.
The Toyota Smarthome Dataset consists of two versions: Trimmed and Untrimmed.
| Toyota Smarthome Trimmed | Toyota Smarthome Untrimmed |
|---|---|
 |
 |
Toyota Smarthome Trimmed
Toyota Smarthome Trimmed has been designed for the activity classification task of 31 activities. The videos were clipped per activity, resulting in a total of 16,115 short RGB+D video samples. activities were performed in a natural manner. As a result, the dataset poses a unique combination of challenges: high intra-class variation, high-class imbalance, and activities with similar motion and high duration variance. Activities were annotated with both coarse and fine-grained labels. These characteristics differentiate Toyota Smarthome Trimmed from other datasets for activity classification.
Directory Structure
📂 Toyota_Smarthome_Trimmed
├── 📁 csvs
│ ├── 📁 cross_subject
│ │ ├── train.csv / test.csv
│ ├── 📁 cross_view_1
│ │ ├── train.csv / test.csv
│ ├── 📁 cross_view_2
│ │ ├── train.csv / test.csv
│ ├── cs_label_mappings.csv
│ ├── cv_label_mappings.csv
├── 📁 raw_data
│ ├── rgb.zip
│ ├── skeletons.zip
├── 📁 cropped_224x224_data
│ ├── rgb.zip
│ ├── skeletons.zip
Descriptions
Toyota_Smarthome_Trimmed/csvs/ - The three dataset splits with train and test CSVs. Each row contains the video filename, skeleton filename, and action label. e.g.,
Pour.Frombottle_p25_r00_v09_c02.mp4,Pour.Frombottle_p25_r00_v09_c02_pose3d.json,0
Toyota_Smarthome_Trimmed/raw_data - Contains the raw videos in 640x480 resolution and corresponding skeletons.
Toyota_Smarthome_Trimmed/cropped_224x224_data - Contains the human-cropped videos in 224x224 resolution and corresponding skeletons. The 2D skeletons have been scaled to match the cropped videos, the 3D skeletons are unchanged.
Toyota Smarthome Untrimmed (TSU)
Toyota Smarthome Untrimmed (TSU) is targeting the activity detection task in long untrimmed videos. Therefore, in TSU, we kept the entire recording when the person is visible. The dataset contains 536 videos with an average duration of 21 mins. Since this dataset is based on the same footage video as Toyota Smarthome Trimmed version, it features the same challenges and introduces additional ones. To better tackle the real-world challenges in the untrimmed video, we densely annotate the dataset with 51 activities.
Directory Structure
📂 Toyota_Smarthome_Untrimmed
├── Annotations.zip
├── Videos_mp4_raw.zip
├── Videos_mp4_cropped224.zip
├── Skeletons.zip
├── 📁 pre-extracted-features
│ ├── Video_mp4-I3D_features.zip
│ ├── Video_mp4-CLIP_features.zip
│ ├── Skeletons-AGCN_features.zip
Data Details
Toyota_Smarthome_Untrimmed/Annotations.zip - Contains the temporally annotated actions for each video. The structure is as follows, each CSV contains the action name, start frame, and end frame:
📂 Annotations
├── 📂 P02
│ ├── P02T01C06.csv
│ ├── P02T02C06.csv
│ ├── ...
├── 📂 P03
├── ...
Toyota_Smarthome_Untrimmed/Videos_mp4_raw.zip - Contains the raw videos in 640x480 resolution.
(Coming soon) Toyota_Smarthome_Untrimmed/Videos_mp4_cropped224.zip - Contains the human-cropped videos in 224x224 resolution.
Toyota_Smarthome_Untrimmed/Skeletons.zip - 2D and 3D skeletons corresponding to the raw videos.
Toyota_Smarthome_Untrimmed/pre-extracted-features - Pre-extracted features for action detection using the cropped version of the videos.
- CLIP features: Frames were extracted from the long videos at 30fps, processed through a pre-trained CLIP-B/32 to obtain frame-wise features, then temporally pooled by averaging every 16 consecutive frame features into a single feature vector
- I3D features: Videos were divided into segments consisting of 16 frames, each segment was processed through a fine-tuned I3D model to obtain a single feature vector per segment (more details available in Section 4.1 of the TSU paper)
- (Coming soon) Skeleton features: The video was divided into segments containing 16 frames and each was processed through a fine-tuned 2s-AGCN model to obtain a single feature vector per segment (more details available in Section 4.1 of the TSU paper)
Citation
Toyota Smarthome Trimmed
@InProceedings{Das_2019_ICCV,
author = {Das, Srijan and Dai, Rui and Koperski, Michal and Minciullo, Luca and Garattoni, Lorenzo and Bremond, Francois and Francesca, Gianpiero},
title = {Toyota Smarthome: Real-World Activities of Daily Living},
booktitle = {The IEEE International Conference on Computer Vision (ICCV)},
month = {October},
year = {2019}
}
Toyota Smarthome Untrimmed
@misc{Dai_2020_arxiv,
author = {Dai, Rui and Das, Srijan and Sharma, Saurav and Minciullo, Luca and Garattoni, Lorenzo and Bremond, Francois and Francesca, Gianpiero},
title = {Toyota Smarthome Untrimmed: Real-World Untrimmed Videos for Activity Detection},
year = {2020},
eprint = {2010.14982},
archivePrefix = {arXiv},
primaryClass = {cs.CV}
}
@ARTICLE{Dai_2022_PAMI,
author={Dai, Rui and Das, Srijan and Sharma, Saurav and Minciullo, Luca and Garattoni, Lorenzo and Bremond, Francois and Francesca, Gianpiero},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={Toyota Smarthome Untrimmed: Real-World Untrimmed Videos for Activity Detection},
year={2022},
volume={},
number={},
pages={1-1},
doi={10.1109/TPAMI.2022.3169976}}
- Downloads last month
- 58