本文主要解析目标检测中常用的COCOAPI工具计算mAP的过程，以及增加相关功能用于更好的提供模型优化的方向。
程序入口
python eval_coco.py
特别说明
results_test.json格式如下：
[{"image_id": 19, "category_id": 1, "bbox": [121.4, 116.02, 560.56, 303.83], "score": 0.97}, {"image_id": 19, "category_id": 1, "bbox": [119.3, 748.22, 566.03, 267.83], "score": 0.95}, .......
{"image_id": 320, "category_id": 3, "bbox": [329.74, 992.53, 35.72, 9.86], "score": 0.62}]
其中image_id和category_id和instances_test.json中的保持一致，而instances_test.json就是标准的coco格式的gt文件。具体格式如下：
{"licenses": [{"name": "", "id": 0, "url": ""}], "info": {"contributor": "", "date_created": "2020-11-16", "description": "table_parse_second_public", "url": "", "version": 2, "year": "2020"}, "categories": [{"id": 1, "name": "bordered", "supercategory": ""}, {"id": 2, "name": "borderless", "supercategory": ""}, {"id": 3, "name": "cell", "supercategory": ""}], "images": [{"coco_url": "", "date_captured": "", "flickr_url": "", "license": 0, "id": 19, "file_name": "cTDaR_t10019.jpg", "height": 1123, "width": 794}, {"coco_url": "", "date_captured": "", "flickr_url": "", "license": 0, "id": 21, "file_name": "cTDaR_t10021.jpg", "height": 1059, "width": 794}, {"coco_url": "", "date_captured": "", "flickr_url": "", "license": 0, "id": 320, "file_name": "cTDaR_t10507.jpg", "height": 1056, "width": 816}],"annotations": [{"category_id": 1, "id": 1218, "image_id": 19, "iscrowd": 0, "segmentation": [[110.0, 96.0, 683.0, 96.0, 683.0, 437.0, 110.0, 437.0]], "area": 195393.0, "bbox": [110.0, 96.0, 573.0, 341.0]}, {"category_id": 1, "id": 1219, "image_id": 19, "iscrowd": 0, "segmentation": [[110.0, 732.0, 683.0, 732.0, 683.0, 1025.0, 110.0, 1025.0]], "area": 167889.0, "bbox": [110.0, 732.0, 573.0, 293.0]},{"category_id": 3, "id": 21911, "image_id": 320, "iscrowd": 0, "segmentation": [[416.0, 845.0, 428.0, 845.0, 428.0, 855.0, 416.0, 855.0]], "area": 120.0, "bbox": [416.0, 845.0, 12.0, 10.0]}]}
eval_coco.py程序：
##The code of eval_coco.py
import pycocotools.coco as coco
from pycocotools.cocoeval import COCOeval
results = r'./results_test.json' ##模型预测结果
anno = r'./instances_test2017.json' ##ground truth
coco_anno = coco.COCO(anno)
coco_dets = coco_anno.loadRes(results)
coco_eval = COCOeval(coco_anno, coco_dets, "bbox")
coco_eval.evaluate()
coco_eval.accumulate()
coco_eval.summarize()
coco_eval.get_good_predict_data()
修改后的cocoeval.py
__author__ = 'tsungyi_ysh'
import numpy as np
import datetime
import time
from collections import defaultdict
from . import mask as maskUtils
import copy
import json
class COCOeval:
# Interface for evaluating detection on the Microsoft COCO dataset.
#
# The usage for CocoEval is as follows:
# cocoGt=..., cocoDt=... # load dataset and results
# E = CocoEval(cocoGt,cocoDt); # initialize CocoEval object
# E.params.recThrs = ...; # set parameters as desired
# E.evaluate(); # run per image evaluation
# E.accumulate(); # accumulate per image results
# E.summarize(); # display summary metrics of results
# For example usage see evalDemo.m and http://mscoco.org/.
#
# The evaluation parameters are as follows (defaults in brackets):
# imgIds - [all] N img ids to use for evaluation
# catIds - [all] K cat ids to use for evaluation
# iouThrs - [.5:.05:.95] T=10 IoU thresholds for evaluation
# recThrs - [0:.01:1] R=101 recall thresholds for evaluation
# areaRng - [...] A=4 object area ranges for evaluation
# maxDets - [1 10 100] M=3 thresholds on max detections per image
# iouType - ['segm'] set iouType to 'segm', 'bbox' or 'keypoints'
# iouType replaced the now DEPRECATED useSegm parameter.
# useCats - [1] if true use category labels for evaluation
# Note: if useCats=0 category labels are ignored as in proposal scoring.
# Note: multiple areaRngs [Ax2] and maxDets [Mx1] can be specified.
#
# evaluate(): evaluates detections on every image and every category and
# concats the results into the "evalImgs" with fields:
# dtIds - [1xD] id for each of the D detections (dt)
# gtIds - [1xG] id for each of the G ground truths (gt)
# dtMatches - [TxD] matching gt id at each IoU or 0
# gtMatches - [TxG] matching dt id at each IoU or 0
# dtScores - [1xD] confidence of each dt
# gtIgnore - [1xG] ignore flag for each gt
# dtIgnore - [TxD] ignore flag for each dt at each IoU
#
# accumulate(): accumulates the per-image, per-category evaluation
# results in "evalImgs" into the dictionary "eval" with fields:
# params - parameters used for evaluation
# date - date evaluation was performed
# counts - [T,R,K,A,M] parameter dimensions (see above)
# precision - [TxRxKxAxM] precision for every evaluation setting
# recall - [TxKxAxM] max recall for every evaluation setting
# Note: precision and recall==-1 for settings with no gt objects.
#
# See also coco, mask, pycocoDemo, pycocoEvalDemo
#
# Microsoft COCO Toolbox. version 2.0
# Data, paper, and tutorials available at: http://mscoco.org/
# Code written by Piotr Dollar and Tsung-Yi Lin, 2015.
# Licensed under the Simplified BSD License [see coco/license.txt]
def __init__(self, cocoGt=None, cocoDt=None, iouType='segm'):
'''
Initialize CocoEval using coco APIs for gt and dt
:param cocoGt: coco object with ground truth annotations
:param cocoDt: coco object with detection results
:return: None
'''
if not iouType:
print('iouType not specified. use default iouType segm')
self.cocoGt = cocoGt # ground truth COCO API
self.cocoDt = cocoDt # detections COCO API
self.evalImgs = defaultdict(list) # per-image per-category evaluation results [KxAxI] elements
self.eval = {} # accumulated evaluation results
self._gts = defaultdict(list) # gt for evaluation
self._dts = defaultdict(list) # dt for evaluation
self.params = Params(iouType=iouType) # parameters
self._paramsEval = {} # parameters for evaluation
self.stats = [] # result summarization
self.ious = {} # ious between all gts and dts
print('-----')
if not cocoGt is None:
self.params.imgIds = sorted(cocoGt.getImgIds())
self.params.catIds = sorted(cocoGt.getCatIds())
print('length of self.params.imgIds:',len(self.params.imgIds))
print('self.params.catIds:',self.params.catIds)
def _prepare(self):
'''
Prepare ._gts and ._dts for evaluation based on params
:return: None
'''
def _toMask(anns, coco):
# modify ann['segmentation'] by reference
for ann in anns:
rle = coco.annToRLE(ann)
ann['segmentation'] = rle
p = self.params
##通过查看保存的hk_noline检测的json，gts是单幅图像100个检测框，类别都是1(因为hk_noline只有一个类别)
##gt是一幅图像对应的gt框，这里的hk_noline是单类别，所以useCats是0，是1，保存的json内容都是一样的
##具体的gts和dts的json格式是一个列表，每一个元素是一个字典，一个字典是一个检测框信息；
##进一步测试下多类的情况，不同的useCats效果？？
if p.useCats:
gts=self.cocoGt.loadAnns(self.cocoGt.getAnnIds(imgIds=p.imgIds, catIds=p.catIds))
dts=self.cocoDt.loadAnns(self.cocoDt.getAnnIds(imgIds=p.imgIds, catIds=p.catIds))
f_gts = open('./tmp1214/gts_catid.json','w+')
json_gt = json.dumps(gts)
f_gts.write(json_gt)
f_gts.close()
f_dts = open('./tmp1214/dts_catid.json','w+')
json_dt = json.dumps(dts)
f_dts.write(json_dt)
f_dts.close()
#print('gts:',gts)
#print('dts:',dts)
else:
gts=self.cocoGt.loadAnns(self.cocoGt.getAnnIds(imgIds=p.imgIds))
dts=self.cocoDt.loadAnns(self.cocoDt.getAnnIds(imgIds=p.imgIds))
f_gts = open('./tmp1214/gts_no_catid.json','w+')
json_gt = json.dumps(gts)
f_gts.write(json_gt)
f_gts.close()
f_dts = open('./tmp1214/dts_no_catid.json','w+')
json_dt = json.dumps(dts)
f_dts.write(json_dt)
f_dts.close()
# convert ground truth to mask if iouType == 'segm'
if p.iouType == 'segm':
_toMask(gts, self.cocoGt)
_toMask(dts, self.cocoDt)
# set ignore flag
for gt in gts:
gt['ignore'] = gt['ignore'] if 'ignore' in gt else 0
gt['ignore'] = 'iscrowd' in gt and gt['iscrowd']
if p.iouType == 'keypoints':
gt['ignore'] = (gt['num_keypoints'] == 0) or gt['ignore']
##这种声明方式产生的self._gts是一个字典，每个元素是列表
##这样得到的就是相同的img_id和类别id的信息，存放在一个列表中，即一张图像的同一个类别的框在一个列表中；
self._gts = defaultdict(list) # gt for evaluation
self._dts = defaultdict(list) # dt for evaluation
## gts中一个gt格式：{"area": 735345, "iscrowd": 0, "image_id": 20190000781, "bbox": [225, 1052, 1257, 585], "category_id": 1, "id": 1063, "ignore": 0, "segmentation": []}
for gt in gts:
self._gts[gt['image_id'], gt['category_id']].append(gt)
## dts中一个dt格式：{"image_id": 20190000781, "category_id": 1, "bbox": [1584.88, 884.44, 152.43, 308.34], "score": 0.0, "segmentation": [[1584.88, 884.44, 1584.88, 1192.78, 1737.3100000000002, 1192.78, 1737.3100000000002, 884.44]], "area": 47000.2662, "id": 78100, "iscrowd": 0}
for dt in dts:
self._dts[dt['image_id'], dt['category_id']].append(dt)
self.evalImgs = defaultdict(list) # per-image per-category evaluation results
self.eval = {} # accumulated evaluation results
self.gt_id2img_id = {}
for gt_i in gts:
self.gt_id2img_id[gt_i['id']] = gt_i['image_id']
self.gt_imgid_cat_id = {}
for gt_i in gts:
if gt_i['image_id'] not in self.gt_imgid_cat_id.keys():
self.gt_imgid_cat_id[gt_i['image_id']] = {}
for cat in self.params.catIds:
self.gt_imgid_cat_id[gt_i['image_id']][cat] = []
self.gt_imgid_cat_id[gt_i['image_id']][gt_i['category_id']].append(gt_i['id'])
#self.gt_imgid_cat_id[gt_i['image_id']][gt_i['category_id']].append(gt_i['id'])
def evaluate(self):
'''
Run per image evaluation on given images and store results (a list of dict) in self.evalImgs
:return: None
'''
tic = time.time()
print('Running per image evaluation...')
p = self.params
# add backward compatibility if useSegm is specified in params
if not p.useSegm is None:
p.iouType = 'segm' if p.useSegm == 1 else 'bbox'
print('useSegm (deprecated) is not None. Running {} evaluation'.format(p.iouType))
print('Evaluate annotation type *{}*'.format(p.iouType))
p.imgIds = list(np.unique(p.imgIds)) ##唯一imgid
if p.useCats:
p.catIds = list(np.unique(p.catIds)) ##唯一gt类别id，不包括背景
p.maxDets = sorted(p.maxDets)
self.params=p
self._prepare()
# loop through images, area range, max detection number
catIds = p.catIds if p.useCats else [-1]
if p.iouType == 'segm' or p.iouType == 'bbox':
computeIoU = self.computeIoU
elif p.iouType == 'keypoints':
computeIoU = self.computeOks
##self.ious是一个字典，每一个元素是表示一张图中某一个类别的预测框(m个)和这个类别的gt(n个)的iou矩阵(m,n)
self.ious = {(imgId, catId): computeIoU(imgId, catId) \
for imgId in p.imgIds
for catId in catIds}
evaluateImg = self.evaluateImg
maxDet = p.maxDets[-1]
##self.evalImgs是列表，每一个元素是字典，存储的是单张图片，一种类别，特定areaRng下的预测框和gt的匹配结果(在不同的阈值下)
self.evalImgs = [evaluateImg(imgId, catId, areaRng, maxDet)
for catId in catIds
for areaRng in p.areaRng
for imgId in p.imgIds
]
self._paramsEval = copy.deepcopy(self.params)
toc = time.time()
print('DONE (t={:0.2f}s).'.format(toc-tic))
def computeIoU(self, imgId, catId):
p = self.params
if p.useCats:
gt = self._gts[imgId,catId]
dt = self._dts[imgId,catId]
else:
gt = [_ for cId in p.catIds for _ in self._gts[imgId,cId]]
dt = [_ for cId in p.catIds for _ in self._dts[imgId,cId]]
if len(gt) == 0 and len(dt) ==0:
return []
##inds是score从大到小排列的索引
inds = np.argsort([-d['score'] for d in dt], kind='mergesort')
##将此处的dt(一张图片一个类别的所有100个检测框(dt大于100个检测框的，按置信度取前100个(100个由p.maxDets设定))按置信度从大到小排列)
##注意是一张图片一种类别的预测框不超过p.maxDets[-1]个，而不是一张图片的预测框不超过这么多，除非设置忽视类别，那就等价于一张图片的总的预测框不多于p.maxDets[-1]
dt = [dt[i] for i in inds]
if len(dt) > p.maxDets[-1]:
dt=dt[0:p.maxDets[-1]]
if p.iouType == 'segm':
g = [g['segmentation'] for g in gt]
d = [d['segmentation'] for d in dt]
elif p.iouType == 'bbox':
g = [g['bbox'] for g in gt]
d = [d['bbox'] for d in dt]
else:
raise Exception('unknown iouType for iou computation')
##gt和dt是一张图片的一种类别的所有框信息；其中dt中只取p.maxDets[-1]个检测框，按置信度从大到小排序；
##g和d是从gt和dt中获取的segmentation信息(分割任务)，检测任务取得是bbox信息；
# compute iou between each dt and gt region
iscrowd = [int(o['iscrowd']) for o in gt]
ious = maskUtils.iou(d,g,iscrowd) ##ious是(m,n),m是d的个数，即模型的预测检测框个数，n是g的框个数
return ious
def computeOks(self, imgId, catId):
p = self.params
# dimention here should be Nxm
gts = self._gts[imgId, catId]
dts = self._dts[imgId, catId]
inds = np.argsort([-d['score'] for d in dts], kind='mergesort')
dts = [dts[i] for i in inds]
if len(dts) > p.maxDets[-1]:
dts = dts[0:p.maxDets[-1]]
# if len(gts) == 0 and len(dts) == 0:
if len(gts) == 0 or len(dts) == 0:
return []
ious = np.zeros((len(dts), len(gts)))
sigmas = p.kpt_oks_sigmas
vars = (sigmas * 2)**2
k = len(sigmas)
# compute oks between each detection and ground truth object
for j, gt in enumerate(gts):
# create bounds for ignore regions(double the gt bbox)
g = np.array(gt['keypoints'])
xg = g[0::3]; yg = g[1::3]; vg = g[2::3]
k1 = np.count_nonzero(vg > 0)
bb = gt['bbox']
x0 = bb[0] - bb[2]; x1 = bb[0] + bb[2] * 2
y0 = bb[1] - bb[3]; y1 = bb[1] + bb[3] * 2
for i, dt in enumerate(dts):
d = np.array(dt['keypoints'])
xd = d[0::3]; yd = d[1::3]
if k1>0:
# measure the per-keypoint distance if keypoints visible
dx = xd - xg
dy = yd - yg
else:
# measure minimum distance to keypoints in (x0,y0) & (x1,y1)
z = np.zeros((k))
dx = np.max((z, x0-xd),axis=0)+np.max((z, xd-x1),axis=0)
dy = np.max((z, y0-yd),axis=0)+np.max((z, yd-y1),axis=0)
e = (dx**2 + dy**2) / vars / (gt['area']+np.spacing(1)) / 2
if k1 > 0:
e=e[vg > 0]
ious[i, j] = np.sum(np.exp(-e)) / e.shape[0]
return ious
def evaluateImg(self, imgId, catId, aRng, maxDet):
'''
perform evaluation for single category and image
:return: dict (single image results)
'''
p = self.params
if p.useCats:
gt = self._gts[imgId,catId]
dt = self._dts[imgId,catId]
else:
gt = [_ for cId in p.catIds for _ in self._gts[imgId,cId]]
dt = [_ for cId in p.catIds for _ in self._dts[imgId,cId]]
if len(gt) == 0 and len(dt) ==0:
return None
for g in gt:
if g['ignore'] or (g['area']<aRng[0] or g['area']>aRng[1]):
g['_ignore'] = 1
else:
g['_ignore'] = 0
# sort dt highest score first, sort gt ignore last
gtind = np.argsort([g['_ignore'] for g in gt], kind='mergesort')
gt = [gt[i] for i in gtind]
dtind = np.argsort([-d['score'] for d in dt], kind='mergesort')
dt = [dt[i] for i in dtind[0:maxDet]]
iscrowd = [int(o['iscrowd']) for o in gt]
# load computed ious
##两种情况，一张图片中，一种类别的gt存在，则
ious = self.ious[imgId, catId][:, gtind] if len(self.ious[imgId, catId]) > 0 else self.ious[imgId, catId]
T = len(p.iouThrs)
G = len(gt)
D = len(dt)
gtm = np.zeros((T,G)) ##存储的是每一个iou阈值、p.maxDet[-1]下的gt能够匹配到的最大iou对应的模型预测框的id，匹配不到的值是0；
dtm = np.zeros((T,D)) ##存储的是每一个iou阈值下的模型预测框匹配到的gt的id，匹配不到的是0；
gtIg = np.array([g['_ignore'] for g in gt])
dtIg = np.zeros((T,D)) ##表示每一个阈值下的预测框匹配到的gt是否需要ignore
##dt已经按照置信度排过序，gt已经按照ignore排过位置，非ignore在前，ignore在后面
##下面的if里面实现的功能是每一个iou阈值下，遍历预测框(预测框已经按置信度从大到小排序)，一个预测框和gt匹配上，则
##另一个预测框不能再通过iou和这个gt进行匹配
if not len(ious)==0:
for tind, t in enumerate(p.iouThrs):
for dind, d in enumerate(dt):
# information about best match so far (m=-1 -> unmatched)
iou = min([t,1-1e-10])
# # 如果m= -1 代表这个dt没有得到匹配 m代表dt匹配的最好的gt的索引下标
m = -1
for gind, g in enumerate(gt):
# if this gt already matched, and not a crowd, continue
if gtm[tind,gind]>0 and not iscrowd[gind]:
continue
# if dt matched to reg gt, and on ignore gt, stop
if m>-1 and gtIg[m]==0 and gtIg[gind]==1:
break
# continue to next gt unless better match made
if ious[dind,gind] < iou:
continue
# if match successful and best so far, store appropriately
iou=ious[dind,gind]
m=gind
# if match made store id of match for both dt and gt
if m ==-1:
continue
dtIg[tind,dind] = gtIg[m] ##对应的能匹配上gt的预测框是否ignore
dtm[tind,dind] = gt[m]['id'] ##dt匹配上的gt的id
gtm[tind,m] = d['id'] ##gt中的框匹配上的预测框的id
# set unmatched detections outside of area range to ignore
##将dtm中没有匹配到gt的预测框，同时预测框的area在指定的aRng范围外，则设置对应的预测框为ignore
a = np.array([d['area']<aRng[0] or d['area']>aRng[1] for d in dt]).reshape((1, len(dt)))
dtIg = np.logical_or(dtIg, np.logical_and(dtm==0, np.repeat(a,T,0)))
# store results for given image and category
return {
'image_id': imgId,
'category_id': catId,
'aRng': aRng, ##aRng范围外的gt和未匹配到gt的预测框但在aRng范围外都是ignore，匹配到gt的预测框在aRng范围外正常计算，不ignore
'maxDet': maxDet, ##这里是p.maxDets[-1]
'dtIds': [d['id'] for d in dt], ##已经排过序的预测框id
'gtIds': [g['id'] for g in gt],
'dtMatches': dtm, ##(T,D) 其中D是已经按置信度排除的bbox
'gtMatches': gtm, ##(T,G) G是按照aRng等信息排序的不ignore在前，ignore在后的gt
'dtScores': [d['score'] for d in dt], ##已经排过序的score
'gtIgnore': gtIg, ##G指的是单张图片特定aRng的gt是否ignore信息
'dtIgnore': dtIg, ##(T,D)
}
def accumulate(self, p = None):
'''
Accumulate per image evaluation results and store the result in self.eval
:param p: input params for evaluation
:return: None
'''
print('Accumulating evaluation results...')
tic = time.time()
if not self.evalImgs:
print('Please run evaluate() first')
# allows input customized parameters
if p is None:
p = self.params
p.catIds = p.catIds if p.useCats == 1 else [-1]
T = len(p.iouThrs) ##设置的iou阈值的个数
R = len(p.recThrs) ##设置的召回的recThrs阈值的个数
K = len(p.catIds) if p.useCats else 1
A = len(p.areaRng)
M = len(p.maxDets)
G_num = 7800 ##设置的该评估用的数据集的gt总数,可以事先通过cvat查看标注的bbox个数，或者自己评估性的设置一个数
precision = -np.ones((T,R,K,A,M)) # -1 for the precision of absent categories ##这个是存储不同的rec值下的p值，相当于存储了pr曲线的采样点
recall = -np.ones((T,K,A,M))
precision_s = -np.ones((T,K,A,M)) ##真实的精确率值
scores = -np.ones((T,R,K,A,M))
DTMatch = -np.ones((T,K,A,G_num,M))
# create dictionary for future indexing
_pe = self._paramsEval
catIds = _pe.catIds if _pe.useCats else [-1]
setK = set(catIds)
setA = set(map(tuple, _pe.areaRng))
setM = set(_pe.maxDets)
setI = set(_pe.imgIds)
# get inds to evaluate
k_list = [n for n, k in enumerate(p.catIds) if k in setK]
m_list = [m for n, m in enumerate(p.maxDets) if m in setM]
a_list = [n for n, a in enumerate(map(lambda x: tuple(x), p.areaRng)) if a in setA]
i_list = [n for n, i in enumerate(p.imgIds) if i in setI]
I0 = len(_pe.imgIds)
A0 = len(_pe.areaRng)
##根据self.evalImgs的存储形式，遍历时最里层是img_id、次外层是aRng、最外层是类别id
'''
self.evalImgs = [evaluateImg(imgId, catId, areaRng, maxDet)
for catId in catIds
for areaRng in p.areaRng
for imgId in p.imgIds
]
'''
# retrieve E at each category, area range, and max number of detections
for k, k0 in enumerate(k_list): ##类别的索引下标遍历
Nk = k0*A0*I0
for a, a0 in enumerate(a_list): ##aRng的遍历
Na = a0*I0
for m, maxDet in enumerate(m_list):
E = [self.evalImgs[Nk + Na + i] for i in i_list]
E = [e for e in E if not e is None]
if len(E) == 0:
continue
##特定类别、特定aRng的所有图片中每一张图片的maxDet个预测框
dtScores = np.concatenate([e['dtScores'][0:maxDet] for e in E])
# different sorting method generates slightly different results.
# mergesort is used to be consistent as Matlab implementation.
inds = np.argsort(-dtScores, kind='mergesort')
##是将特定类别，特定aRng的所有图片的预测框
# (每张图片特定类别、aRng取置信度从大到小的maxDet个框)
#的置信度拉成一位数组，然后再次从大到小排列；
dtScoresSorted = dtScores[inds]
##dtm、dtIg维度是(T,maxDet个数*图片个数)
dtm = np.concatenate([e['dtMatches'][:,0:maxDet] for e in E], axis=1)[:,inds]
dtIg = np.concatenate([e['dtIgnore'][:,0:maxDet] for e in E], axis=1)[:,inds]
gtIg = np.concatenate([e['gtIgnore'] for e in E]) ##gtIg维度是(图片个数，G)
npig = np.count_nonzero(gtIg==0 ) ##gt不ignore的个数
if npig == 0:
continue
##dtm、dtIg维度是(T,maxDet个数*图片个数)
tps = np.logical_and( dtm, np.logical_not(dtIg) )
fps = np.logical_and(np.logical_not(dtm), np.logical_not(dtIg) )
###GT是特定类别、特定aRng、maxDet下所有图片的gt能被预测到的情况
#GT = [self.evalImgs[Nk + Na + i] for i in i_list]
gtmatch_id = tps * dtm ##(T,图片个数*maxDet)
indice_gt = [np.where(i> 0)for i in gtmatch_id]
unique_id = np.array([np.unique(i[indice_gt[p]]) for p,i in enumerate(gtmatch_id)]) #(T,d（不同的iou下，maxDet下的预测框能检测到的gt，所以维度d维度不一样）)
#gtmatch = np.concatenate([e['gtMatches'] for e in GT], axis=1) ##(T,图片个数*G)
#gt_total_num_k_a = gtmatch.shape[1]
#(T,K,A,G_num,M)
for j,id in enumerate(unique_id):
gt_total_num_k_a = len(id)
DTMatch[j,k,a,:gt_total_num_k_a, m] = id ##存储的是预测框能匹配上的gt的id
tp_sum = np.cumsum(tps, axis=1).astype(dtype=np.float)
fp_sum = np.cumsum(fps, axis=1).astype(dtype=np.float)
for t, (tp, fp) in enumerate(zip(tp_sum, fp_sum)):
tp = np.array(tp)
fp = np.array(fp)
nd = len(tp)
rc = tp / npig
pr = tp / (fp+tp+np.spacing(1))
q = np.zeros((R,)) ##特定召回率下的precision值(pr曲线)
ss = np.zeros((R,)) ##特定召回率下的对应的bbox的置信度
if nd:
recall[t,k,a,m] = rc[-1]
precision_s[t,k,a,m] = pr[-1]
else:
recall[t,k,a,m] = 0
precision_s[t,k,a,m] = 0
# numpy is slow without cython optimization for accessing elements
# use python array gets significant speed improvement
pr = pr.tolist(); q = q.tolist()
for i in range(nd-1, 0, -1):
if pr[i] > pr[i-1]:
pr[i-1] = pr[i]
##这里调用的np.searchsorted表示p.recThrs中每一个值能插入到rc中的位置索引，其中rc必须是升序
inds = np.searchsorted(rc, p.recThrs, side='left')
try:
for ri, pi in enumerate(inds):
q[ri] = pr[pi]
ss[ri] = dtScoresSorted[pi]
except:
pass
precision[t,:,k,a,m] = np.array(q)
scores[t,:,k,a,m] = np.array(ss)
self.eval = {
'params': p,
'counts': [T, R, K, A, M],
'date': datetime.datetime.now().strftime('%Y-%m-%d %H:%M:%S'),
'precision': precision, ##(T,R,K,A,M)
'recall': recall, ##(T,K,A,M)
'precision_s': precision_s,
'scores': scores, ##(T,R,K,A,M)
'DTMatch': DTMatch,
}
toc = time.time()
print('DONE (t={:0.2f}s).'.format( toc-tic))
def get_good_predict_data(self):
'''
该函数主要用于得到评估数据集中gt成功预测的图片，相反可以得到gt预测不好的图片用于离线困难数据挑选；
'''
def get_imgid_excellent_predict(save_path, iouThr, areaRng, maxDets, catId):
p = self.params
##(T,K,A,G_num,M)
DTMatch = self.eval['DTMatch']
aind = [i for i, aRng in enumerate(p.areaRngLbl) if aRng == areaRng]
mind = [i for i, mDet in enumerate(p.maxDets) if mDet == maxDets]
cind = [i for i, cat in enumerate(p.catIds) if cat in catId]
if iouThr is not None:
t = np.where(iouThr == p.iouThrs)[0]
DTMatch = DTMatch[t]
##应该分类别计算，不然统一取unique的话，一张图中只要有一种类别的一个gt被检测出来，这张图片后续就会认为是预测较好
##的样本，但是该图片中同一种类的其他gt或者其他类的gt可能完全没检出，因此需要分类别计算
##iou阈值混在一起没问题，或者取特定的iou阈值就可以了
##其实也就是将gt的id和这里的np.unique(DTMatch[:,cind,aind,:,mind])取差集就知道漏检情况了，因为id是指的gt的框的索引
gt_imgid_cat_id = copy.deepcopy(self.gt_imgid_cat_id)
DTMatch = DTMatch[:,cind,aind,:,mind]
#print(DTMatch.shape)
for i,cat in enumerate(catId):
DTMatch_catid = np.unique(DTMatch[:,i,:])
DTMatch_catid = np.delete(DTMatch_catid,0)
for j in DTMatch_catid:
image_id = self.gt_id2img_id[j]
gt_imgid_cat_id[image_id][cat].remove(j)
null_leak_det = [] ##存储完全检测出gt bbox的图片id
for image_id_i in gt_imgid_cat_id.keys():
num_empty = 0
for catId_i in catId:
#for catId_i in gt_imgid_cat_id[image_id_i]:
if len(gt_imgid_cat_id[image_id_i][catId_i]) ==0:
num_empty += 1
#RuntimeError: dictionary changed size during iteration
#gt_imgid_cat_id[image_id_i].pop(catId_i)
#if num_empty == len(self.params.catIds):
if num_empty == len(catId):
null_leak_det.append(image_id_i)
det_gt = np.array([ self.cocoGt.loadImgs(ids=[i])[0]['file_name'] for i in null_leak_det])
det_gt = np.unique(det_gt)
np.savetxt(save_path, det_gt, fmt='%s', delimiter=',')
#json_str = json.dumps(gt_imgid_cat_id)
for img_id_i in null_leak_det:
gt_imgid_cat_id.pop(img_id_i)
#print('length of gt_imgid_cat_id={}, none_leak_det={}, imgIds={}'.format(len(gt_imgid_cat_id.keys()),len(null_leak_det),len(self.params.imgIds)))
json_str = repr(gt_imgid_cat_id)
with open(save_path.replace('.txt','.json').replace('good','leak_det'), 'w') as json_file:
json_file.write(json_str)
leak_det_gt = gt_imgid_cat_id.keys()
leak_det_gt = np.array([ self.cocoGt.loadImgs(ids=[i])[0]['file_name'] for i in leak_det_gt])
leak_det_gt = np.unique(leak_det_gt)
np.savetxt(save_path.replace('good','leak_det'), leak_det_gt, fmt='%s', delimiter=',')
# DTMatch = np.unique(DTMatch[:,cind,aind,:,mind])
# DTMatch = np.delete(DTMatch,0)
# ###检测框对应的gt id和真实的gt的id的差集就是未检测出的gt的框的id
# #DTMatch.tolist().remove(-1.0)
# #print(DTMatch)
# det_gt = np.array([ self.cocoGt.loadImgs(ids=[self.gt_id2img_id[i]])[0]['file_name'] for i in DTMatch])
# det_gt = np.unique(det_gt)
# # with open(save_path,'w+') as file_object:
# # json.dump(DTMatch,file_object)
# np.savetxt(save_path, det_gt, fmt='%s', delimiter=',')
# # f = open(save_path,'w+')
# # for i in DTMatch:
# # f.write(i)
# # f.write('\n')
# # f.close()
#print('save iou={}| areaRng={}| maxDets={}| catId={} to {}'.format(iouThr, areaRng, maxDets, catId, save_path))
save_path = r'./good_predict.txt'
get_imgid_excellent_predict(save_path, iouThr=.5, areaRng='all', maxDets=self.params.maxDets[0], catId=[self.params.catIds[2]])
def summarize(self):
'''
Compute and display summary metrics for evaluation results.
Note this functin can *only* be applied on the default parameter setting
'''
def _summarize( ap=1, iouThr=None, areaRng='all', maxDets=100 , catId=self.params.catIds):
'''
precision = -np.ones((T,R,K,A,M)) # -1 for the precision of absent categories ##这个是存储不同的rec值下的p值，相当于存储了pr曲线的采样点
recall = -np.ones((T,K,A,M))
precision_s = -np.ones((T,K,A,M)) ##真实的精确率值
'''
p = self.params
iStr = ' {:<18} {} @[ IoU={:<9} | area={:>6s} | maxDets={:>3d} ] = {:0.3f}'
# titleStr = 'Average Precision' if ap == 1 else 'Average Recall'
# typeStr = '(AP)' if ap==1 else '(AR)'
#titleStr = 'Average Precision' if ap == 1 else 'Average Recall'
#typeStr = '(AP)' if ap==1 else '(AR)'
iouStr = '{:0.2f}:{:0.2f}'.format(p.iouThrs[0], p.iouThrs[-1]) \
if iouThr is None else '{:0.2f}'.format(iouThr)
aind = [i for i, aRng in enumerate(p.areaRngLbl) if aRng == areaRng]
mind = [i for i, mDet in enumerate(p.maxDets) if mDet == maxDets]
cind = [i for i, cat in enumerate(p.catIds) if cat in catId]
if ap == 1:
titleStr = 'Average P-R curve Area'
typeStr = '(mAP)'
# dimension of precision: [TxRxKxAxM]
s = self.eval['precision']
# IoU
if iouThr is not None:
t = np.where(iouThr == p.iouThrs)[0]
s = s[t]
#s = s[:,:,:,aind,mind]
s = s[:,:,cind,aind,mind]
elif ap == 0:
titleStr = 'Average Recall'
typeStr = '(AR)'
# dimension of recall: [TxKxAxM]
s = self.eval['recall']
if iouThr is not None:
t = np.where(iouThr == p.iouThrs)[0]
s = s[t]
#s = s[:,:,aind,mind]
s = s[:,cind,aind,mind]
else:
titleStr = 'Average Precision'
typeStr = '(AP)'
# dimension of precision: [TxKxAxM]
s = self.eval['precision_s']
# IoU
if iouThr is not None:
t = np.where(iouThr == p.iouThrs)[0]
s = s[t]
#s = s[:,:,aind,mind]
s = s[:,cind,aind,mind]
if len(s[s>-1])==0:
mean_s = -1
else:
mean_s = np.mean(s[s>-1])
print(iStr.format(titleStr, typeStr, iouStr, areaRng, maxDets, mean_s))
return mean_s
def _summarizeDets():
'''
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.902
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.985
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.975
Average Precision (AP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Precision (AP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
Average Precision (AP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.902
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.687
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.932
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.932
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.932
'''
'''
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.902
Average P-R curve Area (mAP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.985
Average P-R curve Area (mAP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.975
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.902
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.687
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.932
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.932
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = -1.000
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.932
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.936
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.127
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.013
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 1 ] = 0.988
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 10 ] = 0.136
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 1 ] = 0.981
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.135
'''
#stats = np.zeros((12,))
stats = np.zeros((31,))
stats[0] = _summarize(1)
stats[1] = _summarize(1, iouThr=.5, maxDets=self.params.maxDets[2])
stats[2] = _summarize(1, iouThr=.75, maxDets=self.params.maxDets[2])
stats[3] = _summarize(1, areaRng='small', maxDets=self.params.maxDets[2])
stats[4] = _summarize(1, areaRng='medium', maxDets=self.params.maxDets[2])
stats[5] = _summarize(1, areaRng='large', maxDets=self.params.maxDets[2])
stats[6] = _summarize(0, iouThr=.5, maxDets=self.params.maxDets[1])
stats[7] = _summarize(0, maxDets=self.params.maxDets[1])
stats[8] = _summarize(0, maxDets=self.params.maxDets[2])
#stats[8] = _summarize(0, iouThr=.5, maxDets=self.params.maxDets[0])
stats[9] = _summarize(0, areaRng='small', maxDets=self.params.maxDets[2])
stats[10] = _summarize(0, areaRng='medium', maxDets=self.params.maxDets[2])
stats[11] = _summarize(0, areaRng='large', maxDets=self.params.maxDets[2])
stats[12] = _summarize(2, maxDets=self.params.maxDets[0])
stats[13] = _summarize(2, maxDets=self.params.maxDets[1])
stats[14] = _summarize(2, maxDets=self.params.maxDets[2])
stats[15] = _summarize(2, iouThr=.5, maxDets=self.params.maxDets[0])
stats[16] = _summarize(2, iouThr=.5, maxDets=self.params.maxDets[1])
stats[17] = _summarize(2, iouThr=.75, maxDets=self.params.maxDets[0])
stats[18] = _summarize(2, iouThr=.75, maxDets=self.params.maxDets[1])
stats[19] = _summarize(2, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[0]])
stats[20] = _summarize(2, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[1]])
stats[21] = _summarize(2, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[2]])
stats[22] = _summarize(0, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[0]])
stats[23] = _summarize(0, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[1]])
stats[24] = _summarize(0, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[2]])
stats[25] = _summarize(1, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[0]])
stats[26] = _summarize(1, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[1]])
stats[27] = _summarize(1, iouThr=.75, maxDets=self.params.maxDets[1], catId=[self.params.catIds[2]])
stats[28] = _summarize(1, areaRng='small', maxDets=self.params.maxDets[2], catId=[self.params.catIds[2]])
stats[29] = _summarize(1, areaRng='medium', maxDets=self.params.maxDets[2], catId=[self.params.catIds[2]])
stats[30] = _summarize(1, areaRng='large', maxDets=self.params.maxDets[2], catId=[self.params.catIds[2]])
# stats = np.zeros((2,))
# stats[0] = _summarize(0, iouThr=.8, maxDets=self.params.maxDets[0])
# stats[1] = _summarize(2, iouThr=.8, maxDets=self.params.maxDets[0])
return stats
def _summarizeKps():
stats = np.zeros((10,))
stats[0] = _summarize(1, maxDets=20)
stats[1] = _summarize(1, maxDets=20, iouThr=.5)
stats[2] = _summarize(1, maxDets=20, iouThr=.75)
stats[3] = _summarize(1, maxDets=20, areaRng='medium')
stats[4] = _summarize(1, maxDets=20, areaRng='large')
stats[5] = _summarize(0, maxDets=20)
stats[6] = _summarize(0, maxDets=20, iouThr=.5)
stats[7] = _summarize(0, maxDets=20, iouThr=.75)
stats[8] = _summarize(0, maxDets=20, areaRng='medium')
stats[9] = _summarize(0, maxDets=20, areaRng='large')
return stats
if not self.eval:
raise Exception('Please run accumulate() first')
iouType = self.params.iouType
if iouType == 'segm' or iouType == 'bbox':
summarize = _summarizeDets
elif iouType == 'keypoints':
summarize = _summarizeKps
self.stats = summarize()
def __str__(self):
self.summarize()
class Params:
'''
Params for coco evaluation api
'''
def setDetParams(self):
self.imgIds = []
self.catIds = []
# np.arange causes trouble. the data point on arange is slightly larger than the true value
#array([0.5 , 0.55, 0.6 , 0.65, 0.7 , 0.75, 0.8 , 0.85, 0.9 , 0.95])
self.iouThrs = np.linspace(.5, 0.95, int(np.round((0.95 - .5) / .05)) + 1, endpoint=True)
self.recThrs = np.linspace(.0, 1.00, int(np.round((1.00 - .0) / .01)) + 1, endpoint=True)
self.maxDets = [1, 10, 100]
self.areaRng = [[0 ** 2, 1e5 ** 2], [0 ** 2, 32 ** 2], [32 ** 2, 96 ** 2], [96 ** 2, 1e5 ** 2]]
self.areaRngLbl = ['all', 'small', 'medium', 'large']
self.useCats = 1
def setKpParams(self):
self.imgIds = []
self.catIds = []
# np.arange causes trouble. the data point on arange is slightly larger than the true value
self.iouThrs = np.linspace(.5, 0.95, int(np.round((0.95 - .5) / .05)) + 1, endpoint=True)
self.recThrs = np.linspace(.0, 1.00, int(np.round((1.00 - .0) / .01)) + 1, endpoint=True)
self.maxDets = [20]
self.areaRng = [[0 ** 2, 1e5 ** 2], [32 ** 2, 96 ** 2], [96 ** 2, 1e5 ** 2]]
self.areaRngLbl = ['all', 'medium', 'large']
self.useCats = 1
self.kpt_oks_sigmas = np.array([.26, .25, .25, .35, .35, .79, .79, .72, .72, .62,.62, 1.07, 1.07, .87, .87, .89, .89])/10.0
def __init__(self, iouType='segm'):
if iouType == 'segm' or iouType == 'bbox':
self.setDetParams()
elif iouType == 'keypoints':
self.setKpParams()
else:
raise Exception('iouType not supported')
self.iouType = iouType
# useSegm is deprecated
self.useSegm = None
新增功能
通过该程序，只需要将任意检测模型的预测输出组织成result_test.json形式，ground truth保存成instances_test.json形式，然后就可以直接调用eval_coco.py进行评估。
增加的功能：1、平均精确率的计算；2、可以指定特定CatID进行指标计算；3、保存指定条件下的检测好的样本和错检、漏检样本的名称;
具体说明：除了cocoapi本身的AP(cocoapi原始程序的AP其实是mAP，而且只能计算所有类的mAP，没有计算指定类别的mAP功能)和AR计算，对官方cocoapi修改后新增真正的AP(Average precision)计算值(新增功能1)，即修改后的cocoapi输出三种指标，AP平均精确率、AR平均召回率，mAP：pr曲线围成的面积；同时可以输出指定类别CatID(新增功能2)、指定aRng(small、medium、large)、指定maxDets(每张图每个类别的最多检测框个数)、指定IOUthr下的三个指标值；
除此之外，为了根据模型预测结果分析得到针对性的模型优化方向，可以根据指定条件（CatID、aRng、maxDets、IOUthr）计算测试集中哪些样本按照指定条件完全检出，并将这些样本名称保存在good_predict.txt文件中，同时将存在漏检或错检的bbox的样本名称和检测结果分别保存在leak_det_predict.txt和leak_det_predict.json中，这样就便于进一步分析模型在哪些测试集样本上表现不佳以及表现不佳的原因，进而可以使用离线数据增强或者其他技术对模型进行针对性优化(新增功能3)！！！
三种类别的新版cocoapi调用结果示例：
loading annotations into memory...
Done (t=0.02s)
creating index...
index created!
Loading and preparing results...
DONE (t=0.03s)
creating index...
index created!
-----
length of self.params.imgIds: 30 ##总共30个样本
self.params.catIds: [1, 2, 3] ##总共三种类别
Running per image evaluation...
Evaluate annotation type *bbox*
DONE (t=9.42s).
Accumulating evaluation results...
DONE (t=0.08s).
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.508
Average P-R curve Area (mAP) @[ IoU=0.50 | area= all | maxDets=100 ] = 0.705
Average P-R curve Area (mAP) @[ IoU=0.75 | area= all | maxDets=100 ] = 0.601
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.298
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.426
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.508
Average Recall (AR) @[ IoU=0.50 | area= all | maxDets= 10 ] = 0.564
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.456
Average Recall (AR) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.557
Average Recall (AR) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.345
Average Recall (AR) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.491
Average Recall (AR) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.542
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 1 ] = 0.421
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets= 10 ] = 0.213
Average Precision (AP) @[ IoU=0.50:0.95 | area= all | maxDets=100 ] = 0.153
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 1 ] = 0.566
Average Precision (AP) @[ IoU=0.50 | area= all | maxDets= 10 ] = 0.305
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 1 ] = 0.487
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.254
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.131 ##指定IOUthr
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.114
Average Precision (AP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.517
Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 10 ] = 1.000
Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.579
Average Recall (AR) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.066
Average P-R curve Area (mAP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.951 ##不同类别catID
Average P-R curve Area (mAP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.543
Average P-R curve Area (mAP) @[ IoU=0.75 | area= all | maxDets= 10 ] = 0.058
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= small | maxDets=100 ] = 0.298 ##不同aRng
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area=medium | maxDets=100 ] = 0.426
Average P-R curve Area (mAP) @[ IoU=0.50:0.95 | area= large | maxDets=100 ] = 0.303
说明：官方的cocoapi计算的mAP是IOU从0.5到0.95，每隔0.05下计算的所有类别的AP的平均值；
具体计算是将模型的预测框先按照每一张图片，每一种类别，按照置信度从大到小得到maxDet个框，然后将测试集中特定类别的所有框按置信度总的排序，继而再对排序后的指定类别的所有框计算tp、fp，然后对所有类别求平均得到mAP(pr曲线的面积)；因此也可以知道pr曲线是成反比例，而且是置信度递减的，置信度越低，recall越高，precision越低。注意maxDet是指一张图片，一种类别容许的最多模型检测框个数，而不是一张图片，所有类别；如果一张图片单类别检测框不足maxDet，也直接去单类别所有检测框，并不会其他补0等操作；cocoapi中矩阵默认-1初始化，因此如果模型检测结果不理想或者测试集中没有满足条件的数据(比如数据集中只有large的物体，计算small物体的评价指标)都可能出现-1的计算结果。
1122
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
