Iam trying to train my own SVM using features extracted using HoG. Now i have one question. should the size of positve image must be 64x128? if the positive pictures are small ,and i can resize them to 64*128. however,the targets are wide.how can i build the positive picture?