Datasets:
File size: 3,099 Bytes
7dc606d 5e20a74 7dc606d 5e20a74 7dc606d 5e20a74 7dc606d 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 5e20a74 838e974 | 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 | ---
language:
- en
license: cc-by-nc-4.0
size_categories:
- 1K<n<10K
task_categories:
- video-classification
pretty_name: MOTOR
tags:
- MOTOR
- driving
- two-wheeler
- multimodal
- eye-tracking
- gaze
- indian-traffic
---
# MOTOR — A Multimodal Dataset for Two-Wheeler Rider Behavior Understanding
[**Project Page**](https://varuniiith.github.io/MOTOR-Dataset/) | [**Paper**](https://huggingface.co/papers/2605.22550) | [**Code**](https://github.com/VarunIIITH/MOTOR)
MOTOR is the first large-scale, multi-view, multimodal resource dedicated to two-wheelers in dense, unstructured traffic. It comprises 1,629 annotated sequences (25+ hours of video data) collected from 16 riders and integrates synchronized front, rear, and helmet videos, rider eye-gaze from wearable trackers, on-road audio, and telemetry (GPS, accelerometer, gyroscope).
## Sample Usage
To use this dataset with the official codebase, follow these steps:
### Setup
```bash
conda create -n motor python=3.10 -y
conda activate motor
pip install -r requirements.txt
```
### Training
The following command trains a three-stream fusion model (RGB + gaze + telemetry) on a Swin3D-T backbone:
```bash
python train.py --config configs/swin3dt_gaze_tele.yaml
```
## Structure
```text
annotations.csv # one row per clip
clips/<ride>/front/<clip>.mp4 # front-facing road view
clips/<ride>/helmet/<clip>.mp4 # helmet-mounted view
clips/<ride>/rear/<clip>.mp4 # rear-facing view
clips/<ride>/eye_tracker/<clip>.mp4 # rider's first-person gaze view
clips/<ride>/telemetry/<clip>.csv # per-frame gyro + GPS for the clip
```
Faces and license plates in all camera videos are blurred for privacy.
Two rides (`14`, `15`) do not include the eye-tracker modality. One ride (`12`) has GPS-only telemetry.
## Annotation columns
`Video ID, Clip ID, Timestamp, Road Type, No. of Lanes, Road Marking, Divider, Traffic Density, Event, Head Pose Direction, Legality`
- **Event:** going straight, lane change, overtake, left/right turn, u-turn, weaving through traffic, obstruction avoidance, distracted driving, stop, violation, near collision
- **Legality:** legal / illegal — based on Indian Motor Vehicle Act and local conventions
## Per-clip telemetry CSV schema
`frame, timestamp_s, gyro_x, gyro_y, gyro_z, gps_lat, gps_lon, gps_alt_m, speed_2d_mps, speed_3d_mps`
Resampled to 30 Hz; `timestamp_s` is clip-local.
## Citation
If you find this work useful, please cite:
```bibtex
@inproceedings{paturkar2026motor,
title={MOTOR: A Multimodal Dataset for Two-Wheeler Rider Behavior Understanding},
author={Paturkar, Varun and others},
booktitle={Proceedings of the IEEE International Conference on Robotics and Automation (ICRA)},
year={2026}
}
```
## License
Released under CC BY-NC 4.0 — non-commercial research use only.
## Note
This release contains the annotated clip-level data. The full uncut raw rides (long-form multi-camera recordings together with their continuous telemetry) will be made publicly available soon. |