@inproceedings{hou2020multiview,
title={Multiview Detection with Feature Perspective Transformation},
author={Hou, Yunzhong and Zheng, Liang and Gould, Stephen},
booktitle={ECCV},
year={2020}
}
The MultiviewX dataset dedicates to multiview synthetic pedestrian detection. Using pedestrian models from PersonX, in Unity, we build a novel synthetic dataset MultiviewX. It follows the WILDTRACK dataset for set-up, annotation, and structure.
The MultiviewX dataset is generated on a 25 meter by 16 meter playground. It has 6 cameras that has overlapping field-of-view. The images in MultiviewX dataset are of high resolution, 1920x1080, and are synchronized. To fully exploit the complementary views, calibrations are also provided in MultiviewX dataset.
Please refer to this link for download.
This repo includes the toolkits and utilities for bulding MultiviewX dataset.
How to's
- download (from link) and copy the 2d/3d bbox annotations into
/matchings. - run the following command.
python run_all.py- done.
For multiview pedestiran detection, MultiviewX follows the same evaluation scheme as Wildtrack with MODA, MODP, precission, and recall. Evaluation toolkit can be found here.
| Method | MODA | MODP | Precision | Recall |
|---|---|---|---|---|
| RCNN & clustering [cite] | 18.7 | 46.4 | 63.5 | 43.9 |
| DeepMCD [cite] | 70.0 | 73.0 | 95.7 | 73.3 |
| Deep-Occlusion [cite] | 76.8 | 59.7 | 97.8 | 70.2 |
| MVDet [cite] [code] | 83.9 | 79.6 | 96.8 | 86.7 |