Pattern Analysis, Statistical Modeling and Computational Learning
lastmod 2018-08-17created 2016-01-266 edits
The PASCAL Network of Excellence has created a distributed institute pioneering principled methods of pattern analysis, statistical modeling, and computational learning as core enabling technologies for multimodal interfaces that are capable of natural and seamless interaction with and among individual human users. The resulting expertise has been applied to problems relevant to both multi-modal interfaces and cognitive systems. PASCAL2 will enable a refocusing of the Institute towards the emerging challenges created by the ever expanding applications of adaptive systems technology and their central role in the development of large scale cognitive systems. Furthermore, the funding will enable the Institute to engage in technology transfer through an Industrial Club to effect rapid deployment of the developed technologies into a wide variety of applications, while undertaking a brokerage of expertise and public outreach programme to communicate the value and relevance of the achieved results.
PASCAL Visual Object Classes
- An introduction to international competitions in computer vision
- The PASCAL Visual Object Classes Homepage
The PASCAL VOC Challenge was an international competition in object class recognition, one of the technologies in computer vision, held annually eight times in total from 2005 to 2012. More precisely, it was a competition on the performance of detecting objects of specific kinds in input images, and a winner is decided for each of the following 20 object classes (the test data set and ground truth can be downloaded).
The PASCAL VOC Challenge is so well known that virtually all of the landmark methods in image object recognition (object detection) in computer vision came out of this competition. The PASCAL VOC Challenge had been led by Mark Everingham, and after his death the challenge ended with the 2012 competition as the last one (a memorial service in his honor was also held at the 2012 competition). However, the evaluation server is said to remain open, so it is still possible to compare the performance of your own algorithm with past winners.
Detection Results
The top rankings are as follows.
- R-FCN, ResNet Ensemble(VOC+COCO) - 88.4%
- HIK_FRCN - 87.9%
- R-FCN, ResNet (VOC+COCO) - 85.0%
- PVANet+ - 84.2%
- Faster RCNN, ResNet (VOC+COCO) - 83.8%
- PVANet+ (compressed) - 83.7%
- SSD512 VGG16 07++12+COCO - 82.2%
- OHEM+FRCN, VGG16, VOC+COCO - 80.1%
- DSSD513_ResNet101_07++12 - 80.0%
- SSD300 VGG16 07++12+COCO - 79.3%