计算方向梯度直方图，基于经典论文Histograms of Oriented Gradients for Human Detec...

Using the Caltech 101 splits provided on the web and an SVM Classifier you should have the following results: Shape 180 Shape180 Shape360 Shape360 (whole image) (roi) (whole image) (roi) l=0 13.20 14.70 16.20 17.57 l=1 33.07 38.56 36.14 44.37 l=2 46.99 58.76 47.25 61.69 l=3 48.43 61.04 49.28 61.63 merge (exhaustive search) 51.76 64.31 54.33 66.27 merge (varma method) 49.21 60.19 50.00 61.04 NOTE: SVM 1-vs-all has been used to learn the level weights when merging all the pyramid levels. ------------------------------------------------ Using the Caltech 256 splits provided on the web and an SVM Classifier you should have the following results: Shape 180 Shape180 Shape360 Shape360 (whole image) (roi) (whole image) (roi) merge (exhaustive search) 16.81 24.28 19.31 27.17 NOTE: again, SVM 1-vs-all has been used to learn the level weights when merging all the pyramid levels.