VisDrone2018-VID Toolkit for Object Detection in Videos
Introduction
This is the documentation of the VisDrone2018 competitions development kit for detection in videos (VID) challenge.
This code library is for research purpose only, which is modified based on the PASCAL VOC [1] and MS-COCO [2] platforms.
The code is tested on the Windows 10 and macOS Sierra 10.12.6 systems, with the Matlab 2013a/2014b/2016b/2017b platforms.
If you have any questions, please contact us (email:tju.drone.vision@gmail.com).
Citation
If you use our toolkit or dataset, please cite our paper as follows:
@article{zhuvisdrone2018,
title={Vision Meets Drones: A Challenge},
author={Zhu, Pengfei and Wen, Longyin and Bian, Xiao and Haibin, Ling and Hu, Qinghua},
journal={arXiv preprint:1804.07437},
year={2018} }
Dataset
For VID competition, there are three sets of data and labels: training data, validation data, and test-challenge data. There is no overlap between the three sets.
Number of snippets
-------------------------------------------------------------------------------------------------
Dataset Training Validation Test-Challenge
-------------------------------------------------------------------------------------------------
Object detection in videos 56 clips 7 clips 16 clips
24,201 frames 2,819 frames 6,333 frames
-------------------------------------------------------------------------------------------------
The challenge requires a participating algorithm to locate the target bounding boxes in each video frame. We focus on ten object categories of interest including pedestrian, person, car, van, bus, truck, motor, bicycle, awning-tricycle, and tricycle. We manually annotate the bounding boxes of different objects and ignored regiones in each video frame. Annotations on the training and validation sets are publicly available.
The link for downloading the data can be obtained by registering for the challenge at
Evaluation Routines
The notes for the folders:
evalVID.m is the main function to evaluate your detector
- modify the dataset path and result path
- use "isSeqDisplay" to display the groundtruth and detections
VID Submission Format
Submission of the results will consist of TXT files with one line per predicted object.It looks as follows:
<frame_index>,<target_id>,<bbox_left>,<bbox_top>,<bbox_width>,<bbox_height>,<score>,<object_category>,<truncation>,<occlusion>
Name Description
----------------------------------------------------------------------------------------------------------------------------------
<frame_index> The frame index of the video frame
<target_id> In the DETECTION result file, the identity of the target should be set to the constant -1.
In the GROUNDTRUTH file, the identity of the target is used to provide the temporal corresponding
relation of the bounding boxes in different frames.
<bbox_left> The x coordinate of the top-left corner of the predicted bounding box
<bbox_top> The y coordinate of the top-left corner of the predicted object bounding box
<bbox_width> The width in pixels of the predicted object bounding box
<bbox_height> The height in pixels of the predicted object bounding box
<score> The score in the DETECTION file indicates the confidence of the predicted bounding box enclosing
an object instance.
The score in GROUNDTRUTH file is set to 1 or 0. 1 indicates the bounding box is considered in
evaluation, while 0 indicates the bounding box will be ignored.
<object_category> The object category indicates the type of annotated object, (i.e., ignored regions (0), pedestrian (1),
people (2), bicycle (3), car (4), van (5), truck (6), tricycle (7), awning-tricycle (8), bus (9), motor (10),
others (11))
<truncation> The score in the DETECTION file should be set to the constant -1.
The score in the GROUNDTRUTH file indicates the degree of object parts appears outside a frame
(i.e., no truncation = 0 (truncation ratio 0%), and partial truncation = 1 (truncation ratio 1% °´ 50%)).
<occlusion> The score in the DETECTION file should be set to the constant -1.
The score in the GROUNDTRUTH file indicates the fraction of objects being occluded
(i.e., no occlusion = 0 (occlusion ratio 0%), partial occlusion = 1 (occlusion ratio 1% °´ 50%),
and heavy occlusion = 2 (occlusion ratio 50% ~ 100%)).
The detections in the ignored regions and labeled as "others" will be not considered in the evaluation. The sample submission of the detector can be found in our website.
References
[1] E. Park, W. Liu, O. Russakovsky, J. Deng, F.-F. Li, and A. Berg, "Large Scale Visual Recognition Challenge 2017", http://imagenet.org/challenges/LSVRC/2017
[2] T. Lin, M. Maire, S. J. Belongie, J. Hays, P. Perona, D. Ramanan, P. Dollar, and C. L. Zitnick, "Microsoft COCO: common objects in context", in ECCV 2014.
Version History
1.0.3 - Jun 29, 2018
The nms function is removed to avoid confusion. The detections in ignored regions are not considered. The bugs in calculating the overlap score are fixed.
1.0.2 - Apr 27, 2018
The nms function is included. The annotations in ignored regions are not considered.
1.0.1 - Apr 25, 2018
Some bugs in the indexes of groundtruth and detection results are fixed. The display function of groundtruth and detection results are included.
1.0.0 - Apr 19, 2018
Initial release.