faster rcnn 基本概念

rpn：region proposal network
IoU：Intersection-over-Union，交集并集之比
GT：ground truth，GT boxes (x1, y1, x2, y2, label)，左上的坐标 和 右下的坐标 + label
im：image
ROI：region of interest
bbox：bounding-box
regression：和one hot的classification其实区别不大，只是target是（可能为0和1以及）其他数字
anchor：一个box，通过从上一个卷积层的结果 滑动得到很多个，每个要和 GT的box 计算bbox_overlap
bbox_overlap：貌似就是IoU，源码如下

def bbox_overlaps( np.ndarray[DTYPE_t, ndim=2] boxes, #anchor的box np.ndarray[DTYPE_t, ndim=2] query_boxes):  #GT的box
    """ Parameters ---------- boxes: (N, 4) ndarray of float #(x1, y1, x2, y2） query_boxes: (K, 4) ndarray of float #(x1, y1, x2, y2） Returns ------- overlaps: (N, K) ndarray of overlap between boxes and query_boxes """
    cdef unsigned int N = boxes.shape[0]  #个数
    cdef unsigned int K = query_boxes.shape[0]  #个数
    cdef np.ndarray[DTYPE_t, ndim=2] overlaps = np.zeros((N, K), dtype=DTYPE)
    cdef DTYPE_t iw, ih, box_area
    cdef DTYPE_t ua
    cdef unsigned int k, n
    for k in range(K):
        box_area = ( #query_boxes的面积
            (query_boxes[k, 2] - query_boxes[k, 0] + 1) *
            (query_boxes[k, 3] - query_boxes[k, 1] + 1)
        )
        for n in range(N):
            iw = (
                min(boxes[n, 2], query_boxes[k, 2]) -
                max(boxes[n, 0], query_boxes[k, 0]) + 1
            )
            if iw > 0:  #两个box的w有交集
                ih = (
                    min(boxes[n, 3], query_boxes[k, 3]) -
                    max(boxes[n, 1], query_boxes[k, 1]) + 1
                )
                if ih > 0:  #两个box的h有交集
                    ua = float( 
                        (boxes[n, 2] - boxes[n, 0] + 1) *
                        (boxes[n, 3] - boxes[n, 1] + 1) +
                        box_area - iw * ih
                    )
                    overlaps[n, k] = iw * ih / ua #交集面积 除以 并集面积
    return overlaps