def __init__(self, y_true_path, y_pred_path, name_path):
# load y_true
self.gt = decode_annotation(y_true_path, type='y_true')
# load y_pred
self.dt = decode_annotation(y_pred_path, type='y_pred')
self.names = decode_name(name_path)
def __init__(self, cfg, verbose=0):
self.verbose = verbose
self.mask = cfg["yolo"]["mask"]
self.anchors = cfg["yolo"]["anchors"]
self.max_boxes = cfg["yolo"]["max_boxes"]
self.strides = cfg["yolo"]["strides"]
self.anno_path = cfg["train"]["anno_path"]
self.name_path = cfg['train']['name_path']
self.image_size = cfg["train"]["image_size"]
self.batch_size = cfg["train"]["batch_size"]
self.mix_up = cfg['train']["mix_up"]
self.cut_mix = cfg['train']['cut_mix']
self.mosaic = cfg['train']['mosaic']
self.label_smoothing = cfg['train']["label_smoothing"]
self.annotation = decode_annotation(anno_path=self.anno_path)
self.num_anno = len(self.annotation)
self.name = decode_name(name_path=self.name_path)
self.num_classes = len(self.name)
# init
self._image_size = np.random.choice(self.image_size)
self._anchors = self.anchors / self._image_size
self._grid_size = self._image_size // self.strides
# -*- coding: utf-8 -*-
import numpy as np
import cv2
from core.utils import decode_annotation
# Parameters
K = 6
image_size = 416
dataset_path = './data/pascal_voc/train.txt'
print('Num of Clusters is', K)
print('Base Image Size is', image_size)
# Read Dataset
anns = decode_annotation(dataset_path)
def resize_bboxes(path, bboxes):
image = cv2.imread(path)
h, w, _ = image.shape
scale = min(image_size / w, image_size / h)
nw, nh = int(scale * w), int(scale * h)
dw, dh = (image_size - nw) // 2, (image_size - nh) // 2
bboxes = np.asarray(bboxes).astype(np.float32)
bboxes[:, [0, 2]] = bboxes[:, [0, 2]] * scale + dw
bboxes[:, [1, 3]] = bboxes[:, [1, 3]] * scale + dh
bboxes[:, [0, 2]] = np.clip(bboxes[:, [0, 2]], 0, image_size - 1)
bboxes[:, [1, 3]] = np.clip(bboxes[:, [1, 3]], 0, image_size - 1)
def VOCEval(y_true_path,
y_pred_path,
name_path,
ovthresh=0.5,
use_07_metric=False,
verbose=0):
"""
:param y_true_path:
:param y_pred_path:
:param ovthresh: Overlap threshold (default = 0.5)
:param use_07_metric: Whether to use VOC07's 11 point AP computation (default False)
:return:
"""
# assumes detections are in detpath.format(classname)
# assumes annotations are in annopath.format(imagename)
# assumes imagesetfile is a text file with each line an image name
# cachedir caches the annotations in a pickle file
# load y_true
y_true = decode_annotation(y_true_path, type='y_true')
# load y_pred
y_pred = decode_annotation(y_pred_path, type='y_pred')
names = decode_name(name_path)
ans = {}
for classname_idx in range(len(names)):
# extract gt objects for this class
class_recs = {}
npos = 0
for imgs_path, bboxes, labels in y_true:
image_idx = os.path.basename(imgs_path).split('.')[0]
bbox = [bbox for bbox, label in zip(bboxes, labels) if label == classname_idx]
bbox = np.array(bbox)
det = [False] * len(bbox)
diff = [False] * len(bbox)
npos += len(bbox)
class_recs[image_idx] = {'bbox': bbox,
'det': det,
'difficult': diff}
# extract pd objects for this class
image_ids = []
BB = []
confidence = []
for imgs_path, bboxes, labels, confis in y_pred:
image_idx = os.path.basename(imgs_path).split('.')[0]
for bbox, label, confi in zip(bboxes, labels, confis):
if label != classname_idx:
continue
image_ids.append(image_idx)
BB.append(bbox)
confidence.append(confi)
image_ids = np.array(image_ids)
BB = np.array(BB)
confidence = np.array(confidence)
# sort by confidence
sorted_ind = np.argsort(-confidence)
BB = BB[sorted_ind]
image_ids = [image_ids[x] for x in sorted_ind]
# go down dets and mark TPs and FPs
nd = len(image_ids)
tp = np.zeros(nd)
fp = np.zeros(nd)
for d in range(nd):
R = class_recs[image_ids[d]]
bb = BB[d].astype(float)
ovmax = -np.inf
BBGT = R['bbox'].astype(float)
if BBGT.size > 0:
# compute overlaps
# intersection
ixmin = np.maximum(BBGT[:, 0], bb[0])
iymin = np.maximum(BBGT[:, 1], bb[1])
ixmax = np.minimum(BBGT[:, 2], bb[2])
iymax = np.minimum(BBGT[:, 3], bb[3])
iw = np.maximum(ixmax - ixmin + 1., 0.)
ih = np.maximum(iymax - iymin + 1., 0.)
inters = iw * ih
# union
uni = ((bb[2] - bb[0] + 1.) * (bb[3] - bb[1] + 1.) +
(BBGT[:, 2] - BBGT[:, 0] + 1.) * (BBGT[:, 3] - BBGT[:, 1] + 1.)
- inters)
overlaps = inters / uni
ovmax = np.max(overlaps)
jmax = np.argmax(overlaps)
if ovmax > ovthresh:
if not R['difficult'][jmax]:
if not R['det'][jmax]:
tp[d] = 1.
R['det'][jmax] = 1
else:
fp[d] = 1.
else:
fp[d] = 1.
# compute precision recall
fp = np.cumsum(fp)
tp = np.cumsum(tp)
rec = tp / np.maximum(float(npos), np.finfo(np.float64).eps)
# 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)
ans[names[classname_idx]] = {'rec': rec, 'prec': prec, 'ap': ap}
if verbose > 0:
print("\nOn Test Data")
print("class |AP@50")
mAP = []
for key in ans.keys():
ap = ans[key]['ap']
mAP.append(ap)
if verbose > 0:
print("{:>15}|{:>15.2%}".format(key, ap))
mAP = np.mean(mAP)
if verbose > 0:
print("{:>15}|{:>15.2%}".format('mAP', mAP))
return mAP