Research | Teaching


1. Retail50K Dataset (download)
Retail50K is a collection of 47,000 images from different supermarkets. Annotations on those images are the layer edges of shelves, fridges and displays.
2. PIoU Loss (download)
3. Industrial10 Dataset (download)
Industrial10 is specially designed for evaluating camera pose estimation in industrial environments. For this, we collected a dataset containing 10 industrial objects using our proposed WatchPose approach. The baseline can be found in our paper.
4. MPEG400 Dataset (download)
MPEG400 dataset is a subset of MPEG7 dataset, consisting of 400 objects categorised in 20 classes. Comparing to MPEG7, shapes in MPEG400 have much larger intra-class variations and inter-class similarities.
5. Tetrapod120 Dataset (download)
Tetrapod120 database includes 120 visually similar tetrapod animals with 6 classes, such as camel, cattle, deer, dog, elephant and horse. We aim to evaluate the ability of matching methods for fine-grained shapes where some species are really difficult to distinguish (e.g. horses and dogs).
6. ETHZ Open Curve Dataset (download)
ETHZ is a dataset for testing object class detection algorithms. It contains 255 test images and features five diverse shape-based classes (apple logos, bottles, giraffes, mugs and swans). Based on the shapes from ground truth, we manually generate the following open curves by keeping their contour parts in the parentheses: 44 open curves for apple logos (the right half part), 55 open curves for bottles (the right half part), 91 open curves for giraffes (the upper neck part), 48 open curves for mugs (the right half part with handle) and 32 open curves for swans (the upper half part with head). In total, there are 271 open curves. This dataset is used for evaluating the open curve matching.
7. MPEG7 Contour Segment (CS) Dataset (MPEG7CS, MPEG7CS-Small)
MPEG7 dataset is a standard and commonly used shape dataset for evaluating shape matching and classification. The total number of images in the MPEG7 database is 1400: 70 classes of various shapes, each class with 20 images. We create the MPEG7 CS dataset using its shape contours. For the shapes from the same class in MPEG7, we first extract their contours and then manually remove the same part from contours. Finally, the MPEG7 CS dataset is generated with 1400 CSs. This dataset can be used for evaluating CS descriptors.
8. Source code for parsing /proc metrics (download)
C++, Linux, Copyright © Cong Yang
9. Source code of KidPating in KidSpark (download)
C#, Windows, Copyright © Cong Yang
10. Source code for shape interesting point detection (DCE Method, Shape Trend Method)
These scripts can be used for detection shape boundary interesting points.
Matlab, Copyright © Cong Yang & Christian Feinen
11. Source code for skeleton pruning (download)
Matlab, Copyright © Cong Yang
12. Source code for skeleton graph generation (download)
This script can be used for detecting skeleton endpoints, junction points, skeleton pathes, etc.
Matlab, Copyright © Cong Yang
13. Source code for pairwise and third-order graph matching (download)
These scripts are applied based on the methods in following papers:

[1] M. Leordeanu et al. A spectral technique for correspondence problems using pairwise constraints. In IEEE International Conference on Computer Vision, pages 1482–1489. 2005.
[2] O. Duchenne, et al. A tensor-based algorithm for high-order graph matching. IEEE Transactions on Pattern Analysis and Machine Intelligence, 33(12):2383–2395, 2011.

Matlab, Copyright © Cong Yang
14. Source code for generating contour segment descriptors (download)
Matlab, Copyright © Cong Yang
15. Source code for audio skeleton generation and matching (download)
Matlab, Copyright © Cong Yang