关于coco-api评估的一些小操作

关于cocoapi评估时显示特定类别的一些操作

指定一类输出的时候

在cocoapi-master/PythonAPI中构建evaluate.py

构建evaluate.py的内容如下：

"""
COCO-Style Evaluations
"""
 
import json
import os
 
import argparse
from pycocotools.coco import COCO
from pycocotools.cocoeval import COCOeval
 
def eval(coco_gt, image_ids, pred_json_path):
    # load results in COCO evaluation tool
    coco_pred = coco_gt.loadRes(pred_json_path)
 
    # run COCO evaluation
    print('BBox')
    coco_eval = COCOeval(coco_gt, coco_pred, 'bbox')
    coco_eval.params.imgIds = image_ids
    coco_eval.params.catIds = [2] # 重点在这句！！！！！ ！
    #2代表的是该类对应的coco数据集上的标签序号，根据现实情况修改即可
    coco_eval.evaluate()
    coco_eval.accumulate()
    coco_eval.summarize()
 
if __name__ == '__main__':
    ap = argparse.ArgumentParser()
    ap.add_argument('--gt-json', type=str, default='instances.json', help='coco val2017 annotations json files')
    ap.add_argument('--pred-json', type=str, default='coco_results.json', help='pred coco val2017 annotations json files')
    args = ap.parse_args()
    print(args)
 
    pred_json_path = args.pred_json
 
    MAX_IMAGES = 1000
    coco_gt = COCO(args.gt_json)
    image_ids = coco_gt.getImgIds()[:MAX_IMAGES]
 
    eval(coco_gt, image_ids, pred_json_path)

运行时输入如下命令：

python3 evaluate.py --gt-json  [真实标签json路径]  --pred-json [预测标签json路径]

指定多类输出的时候

在pycocotools/coco.py更改init部分，具体见：

with open(annotation_file, 'r') as f:
                dataset = json.load(f)
            assert type(dataset)==dict, 'annotation file format {} not supported'.format(type(dataset))
            print('Done (t={:0.2f}s)'.format(time.time()- tic))
            print(
                "category names: {}".format([e["name"] for e in sorted(dataset["categories"], key=lambda x: x["id"])]))   # 重点是这句！！！

然后修改pycocotools/cocoeval.py，具体修改内容为：

            #print(iStr.format(titleStr, typeStr, iouStr, areaRng, maxDets, mean_s))
            category_dimension = 1 + int(ap)
            if s.shape[category_dimension] > 1:
 
                iStr += ", per category = {}"
                mean_axis = (0,)
                if ap == 1:
                    mean_axis = (0, 1)
                per_category_mean_s = np.mean(s, axis=mean_axis).flatten()
                with np.printoptions(precision=3, suppress=True, sign=" ", floatmode="fixed"):
                    print(iStr.format(titleStr, typeStr, iouStr, areaRng, maxDets, mean_s, per_category_mean_s))  
            else:
                print(iStr.format(titleStr, typeStr, iouStr, areaRng, maxDets, mean_s, ""))

通过这个修改之后，运行evaluate.py，会按照[类别1，类别2，…,类别n]的形式打印出各AP和AR值。