一面一个教育公司

目标检测有哪些算法，RRPN
单阶段、多阶段

分类损失函数有哪些、回归损失函数

最近看的一篇论文

平时怎么安排工作

手写NMS。哎！当时没有写出来，面试官还说我iou算的过程不对，我仔细一想，确实不对，留下了不学无术的泪水。

import numpy as np
def nms(dets,score):
    x1 = dets[:,0]
    y1 = dets[:,1]
    x2 = dets[:,2]
    y2 = dets[:,3]
    area = (x2-x1+1)*(y2-y1+1)
    scores = dets[:,4]
    idx = scores.argsort()[::-1]
    keep = []
    while idx.size > 0:
        i = idx[0]
        keep.append(i)
        xx1 = np.maximum(x1[i],x1[idx[1:]])
        xx2 = np.minimum(x2[i],x2[idx[1:]])
        yy1 = np.maximum(y1[i],y1[idx[1:]])
        yy2 = np.minimum(y2[i],y2[idx[1:]])
        w = np.maximum(xx2-xx1,0)
        h = np.maximum(yy2-yy1,0)
        overlap = w*h
        iou = overlap/(area[i]+area[idx[1:]]-overlap)
        idx_i = np.where(iou<=score_thr)[0]
        idx = idx[idx_i+1]
    return keep

建议拓宽知识面分类、检测、分割等