YOLOX输出Precision、Recall指标（VOC数据集）

按照yolov5中计算precision和recall的方式，计算最大f1分数时的iou下的p和r

需要修改两个地方：

1、yolox/evaluators/voc_eval.py

在最后几行，计算最大f1分数时的iou下的p和r，并且返回值

        # compute precision recall
    fp = np.cumsum(fp)
    tp = np.cumsum(tp)
    rec = tp / float(npos)
    # avoid divide by zero in case the first detection matches a difficult
    # ground truth
    prec = tp / np.maximum(tp + fp, np.finfo(np.float64).eps)
    ap = voc_ap(rec, prec, use_07_metric)
    #计算最大f1分数时，对应iou的p和r
    f1 = 2 * prec * rec / (prec + rec + 1e-16)
    i = f1.argmax()
    f1_best = f1[i]
    rec_best = rec[i]
    prec_best = prec[i]

    return rec, prec, ap, f1_best, rec_best, prec_best

2、yolox/data/datasets/voc.py

将第一步的返回值输出打印就行了。输出每个类别的P、R以及AP，和所有类别的P、R、mAP。想输出别的指标比如f1、acc等，同理。

    def _do_python_eval(self, output_dir="output", iou=0.5):
        rootpath = os.path.join(self.root, "VOC" + self._year)
        name = self.image_set[0][1]
        annopath = os.path.join(rootpath, "Annotations", "{:s}.xml")
        imagesetfile = os.path.join(rootpath, "ImageSets", "Main", name + ".txt")
        cachedir = os.path.join(
            self.root, "annotations_cache", "VOC" + self._year, name
        )
        if not os.path.exists(cachedir):
            os.makedirs(cachedir)
        aps = []
        recs = []
        precs = []
        # The PASCAL VOC metric changed in 2010
        use_07_metric = True if int(self._year) < 2010 else False
        print("Eval IoU : {:.2f}".format(iou))
        if output_dir is not None and not os.path.isdir(output_dir):
            os.mkdir(output_dir)
        for i, cls in enumerate(VOC_CLASSES):

            if cls == "__background__":
                continue

            filename = self._get_voc_results_file_template().format(cls)
            rec, prec, ap, f1_best, rec_best, prec_best = voc_eval(
                filename,
                annopath,
                imagesetfile,
                cls,
                cachedir,
                ovthresh=iou,
                use_07_metric=use_07_metric,
            )
            aps += [ap]
            recs += [rec_best]
            precs += [prec_best]
            if iou == 0.5:
                print("--------------------------------------------------------------")
                print("AP for {} = {:.4f}".format(cls, ap))
                print("Recall for {}= {:.4f}".format(cls, rec_best))
                print("Precision for {} = {:.4f}".format(cls, prec_best))
            if output_dir is not None:
                with open(os.path.join(output_dir, cls + "_pr.pkl"), "wb") as f:
                    pickle.dump({"rec": rec, "prec": prec, "ap": ap}, f)
        if iou == 0.5:
            print("--------------------------------------------------------------")
            print("Mean AP = {:.4f}".format(np.mean(aps)))
            print("Mean Recall = {:.4f}".format(np.mean(recs)))
            print("Mean Precision = {:.4f}".format(np.mean(precs)))
            # print("~~~~~~~~")
            # print("Results:")
            # for ap in aps:
            #     print("{:.3f}".format(ap))
            # print("{:.3f}".format(np.mean(aps)))
            # print("~~~~~~~~")
            print("")
            print("--------------------------------------------------------------")
            print("Results computed with the **unofficial** Python eval code.")
            print("Results should be very close to the official MATLAB eval code.")
            print("Recompute with `./tools/reval.py --matlab ...` for your paper.")
            print("-- Thanks, The Management")
            print("--------------------------------------------------------------")

        return np.mean(aps)

PS：我是按照yolov5输出precision和recall的计算方法改的，方便和yolov5做对比。如有不正确请各位大佬指正。