Datasets
Pri-SDL Dataset of Object Detection in Aerial Images

UCAS-High Resolution Aerial Object Detection Dataset,which contains three parts.

Car set:510 images

Plane set:1000 images

Negative set:910 images

Each of the images has a ground-truth file with .txt format.

DownLoad: Dataset Instruction Instruction-cn

References:

[1]H. Zhu, X. Chen, W. Dai, K. Fu, Q. Ye, J. Jiao, "Orientation Robust Object Detection in Aerial Images Using Deep Convolutional Neural Network," IEEE Int'l Conf. Image Processing, 2015.[pdf]

Pri-SDL Dataset of Human Detection

We prepare a dataset for human detection, which contains two test sets and three training sets.

Test Set:

SDL_Test_setA 140 images, which is mainly for humans of frontal view.

SDL_Test_setB 258 images,which is mainly for multi-view and multi-posture humans. Some of the images are from human dataset.

Each of the images has a ground-truth file with .txt format.Here is the Groundtruth file format.

Training Sets:

SDL_Training Set frontal view 1000 images

SDL_Training Set side view 3050 images

SDL_Multi-view/posture positives 7550 images

All the images are in BITMAP format, have a size of 64x128 pixels, with a defined bound. For the usage of the datasets, please refer to the follows.

References:

[1]R. Xu, J. Jiao*, B. Zhang, Q. Ye, “Pedestrian Detection in Images via Cascaded L1-Norm Minimization Learning Method,” Pattern Recognition, vol.45, 2573-2583, 2012.[pdf]

[2]Q. Ye*, J. Liang, J. Jiao, "Pedestrian Detection In Video Images Via Error Correcting Output Code Classification Of Manifold Subclasses," IEEE Transactions on Intelligent Transportation Systems, vol.1, pp.193-202, 2012.[pdf]

For more human detection datasets and the evaluation protocol, please refer to the Caltech Pedestrian Detection Benchmark.

Pri-SDL Dataset of Car Detection

Test Set:

Car Test Set(164 images,101M). The images are in BITMAP format. Each image has a ground-truth file with .txt format. Here is the Groundtruth file format.

Car Training Positives  (1528 images, 13.92M) are marked image blobs of BITMAP format. Negatives are not available. We suggest mining negatives from images without vehicles.

For the car evaluation protocol, please refer to the Caltech Pedestrian Detection Benchmark.

Pri-SDL Dataset of Visual Object Tracking

The tracking test video is captured outdoor with complex background and an hand-hold camera. All the video are in AVI format.

Original video (about 60M): video-1 video-2

For the evaluation protocol, please refer to the reference.

Reference: Z. Han, J. Jiao, B. Zhang ,Q. Ye, and J. Liu, "Visual Object Tracking via Sample-Based Adaptive Sparse Representation (AdaSR)," Pattern Recognition, no.44, pp.2170–2183, 2011.[pdf]

Pri-SDL Dataset of Multi-Focus Image Fusion

Multi-Focus Dataset (about 100M) is prepared for the subjective and objective evaluation of multi-focus image fusion algorithms. All the images in the dataset are in JPEG format.

There are totally 12 groups of images for objective evaluation and 15 groups for subjective evaluation. Each group of images for objective evaluation has a ground-truth image, which is the only one color image in the file folder. Other de-focus images in the same folder are in grey.  

When perform objective evaluation, the Root Mean Squared Error (RMSE) and Structural Similarity (SSIM) protocols are employed.

Pri-SDL Dataset of Weld Line Detection and Tracking

We prepare a dataset for weld line detection and tracking, which contains 11 test sets.

Test Set:

Set 1,Set 4,Set 5,Set 6,Set 7and Set 8 are captured in Fenghua, including 648, 45, 200, 483, 168 and 76 frames, respectively.

Set 2 and Set 3are captured in Guanting wind farm of Beijing, including 516 and 556 frames, respectively.

Set 9and Set 10 are captured in SDL(simulation environment), including 727 and 658 frames, respectively..

Set 11contains 28 images for camera and laser sensor calibration.

References:

[1] L. Zhang, J. Jiao, Q. Ye, Z. Han, and W. Yang, "Robust Weld Line Detection with Cross Structured Light and Hidden Markov Model," Proc. of 2012 IEEE International Conference on Mechatronics and Automation,pp.1411-1416,2012.[pdf]

[2] L. Zhang, Q. Ye, W. Yang, J. Jiao, “Weld Line Detection and Tracking via Spatial-Temporal Cascaded Hidden Markov Models and Cross Structured Light” IEEE Trans. Instrumentation and Measurement, Vol.6, No.4, pp.742-752, 2014. [pdf]

Source Codes

Code for Color Image Segmentation with C++ language. The program should be compiled with a Visual Studio C++ later than the version 6.0.

Reference: Q. Ye, W. Gao, T. Huang and W. Wang, "A Color Image Segmentation Algorithm Using Color and Spatial Information,"Chinese Journal of Software, 2004, NO.12.

Code for Text detection in image and video frames with C++ language. The program should be compiled with a Visual Studio C++ later than the version 6.0.

Reference: Q. Ye, Q. Huang, W. Gao, D. Zhao, "Fast and robust text detection in image and video frames," Image and Vision Computing, Vol.23, No.6, pp565-576, Mar.2005.[pdf]

Demo Code with a Readme file and an >Executable file (with ICDAR’11 test images) for Scene Text Detection with C++ language. The program should be compiled with a Visual Studio C++ later than the version 6bushi .0. The experimental result has been uploaded to ICDAR’2013 "Robust Reading" competition.

Reference: Q. Ye and David Doermann, "Scene Text Detection via Integrated Discrimination of Component Appearance and Consensus," Book title: Camera-Based Document Analysis and Recognition, Springer 2014.

Code for visual weld line detection and tracking with C++ language. The program should be compiled with a Visual Studio C++ later than the version 6.0.

Reference: L. Zhang, Q. Ye, W. Yang, J. Jiao, “Weld Line Detection and Tracking via Spatial-Temporal Cascaded Hidden Markov Models and Cross Structured Light” IEEE Trans. Instrumentation and Measurement, Vol.6, No.4, pp.742-752, 2014. [pdf]

Code for pedestrian detection using PCA-CCF features, in Matlab.

Reference: W. Ke, Y. Zhang, P. Wei, Q. Ye, J. Jiao, “Pedestrian Detection via PCA Filters Based Convolutional Channel Features”, IEEE Int'l Conf. Acoustic, Speech and Signal Processing (ICASSP), 2015.[pdf]

Code for WSOD with correlation and part suppression, in Matlab

Reference: F. Wan, P. Wei, Z. Han, K. Fu, Q. Ye, "Weakly Supervised Object Detection with correlation and part suppression", ICIP 2016 (submitted)

Acknowledgements

This datasets and codes belong to University of Chinese Academy of Sciences, with the mentoring support from X. Zhang. We would like to acknowledge the help of several volunteers who annotated these datasets. In particular, we would like to thank Z. Han, N. Zhang, J. Liang, Y. Zhu and Q. Ye for their preparing the datasets and codes.

Main contact: zhangxiaodan10 at mails.ucas.ac.cn

Last modified: Mon. July 06 2015