def __init__(self,
data_dir,
device,
transform,
set_type='test',
year='2007',
display=False):
self.data_dir = data_dir
self.device = device
self.transform = transform
self.labelmap = VOC_CLASSES
self.set_type = set_type
self.year = year
self.display = display
# path
self.devkit_path = os.path.join(data_dir, 'VOC' + year)
self.annopath = os.path.join(data_dir, 'VOC2007', 'Annotations',
'%s.xml')
self.imgpath = os.path.join(data_dir, 'VOC2007', 'JPEGImages',
'%s.jpg')
self.imgsetpath = os.path.join(data_dir, 'VOC2007', 'ImageSets',
'Main', set_type + '.txt')
self.output_dir = self.get_output_dir('voc_eval/', self.set_type)
# dataset
self.dataset = VOCDetection(data_dir=data_dir,
image_sets=[('2007', set_type)],
transform=transform)
def build_dataset(args, train_size, val_size, device):
if args.dataset == 'voc':
data_dir = os.path.join(args.root, 'VOCdevkit')
num_classes = 20
dataset = VOCDetection(data_dir=data_dir,
img_size=train_size,
transform=TrainTransforms(train_size),
color_augment=ColorTransforms(train_size),
mosaic=args.mosaic)
evaluator = VOCAPIEvaluator(data_dir=data_dir,
img_size=val_size,
device=device,
transform=ValTransforms(val_size))
elif args.dataset == 'coco':
data_dir = os.path.join(args.root, 'COCO')
num_classes = 80
dataset = COCODataset(data_dir=data_dir,
img_size=train_size,
image_set='train2017',
transform=TrainTransforms(train_size),
color_augment=ColorTransforms(train_size),
mosaic=args.mosaic)
evaluator = COCOAPIEvaluator(data_dir=data_dir,
img_size=val_size,
device=device,
transform=ValTransforms(val_size))
else:
print('unknow dataset !! Only support voc and coco !!')
exit(0)
return dataset, evaluator, num_classes
print("w, h: ", round(centroid.w, 2), round(centroid.h, 2), "area: ",
round(centroid.w, 2) * round(centroid.h, 2))
return centroids
if __name__ == "__main__":
n_anchors = args.num_anchorbox
img_size = args.img_size
dataset = args.dataset
loss_convergence = 1e-6
iters_n = 1000
dataset_voc = VOCDetection(data_dir=os.path.join(args.root, 'VOCdevkit'),
img_size=img_size)
dataset_coco = COCODataset(data_dir=os.path.join(args.root, 'COCO'),
img_size=img_size)
boxes = []
print("The dataset size: ", len(dataset))
print("Loading the dataset ...")
# VOC
for i in range(len(dataset_voc)):
if i % 5000 == 0:
print('Loading voc data [%d / %d]' % (i + 1, len(dataset_voc)))
# For VOC
img, _ = dataset_voc.pull_image(i)
w, h = img.shape[1], img.shape[0]
class VOCAPIEvaluator():
""" VOC AP Evaluation class """
def __init__(self,
data_dir,
device,
transform,
set_type='test',
year='2007',
display=False):
self.data_dir = data_dir
self.device = device
self.transform = transform
self.labelmap = VOC_CLASSES
self.set_type = set_type
self.year = year
self.display = display
# path
self.devkit_path = os.path.join(data_dir, 'VOC' + year)
self.annopath = os.path.join(data_dir, 'VOC2007', 'Annotations',
'%s.xml')
self.imgpath = os.path.join(data_dir, 'VOC2007', 'JPEGImages',
'%s.jpg')
self.imgsetpath = os.path.join(data_dir, 'VOC2007', 'ImageSets',
'Main', set_type + '.txt')
self.output_dir = self.get_output_dir('voc_eval/', self.set_type)
# dataset
self.dataset = VOCDetection(data_dir=data_dir,
image_sets=[('2007', set_type)],
transform=transform)
def evaluate(self, net):
net.eval()
num_images = len(self.dataset)
# all detections are collected into:
# all_boxes[cls][image] = N x 5 array of detections in
# (x1, y1, x2, y2, score)
self.all_boxes = [[[] for _ in range(num_images)]
for _ in range(len(self.labelmap))]
# timers
det_file = os.path.join(self.output_dir, 'detections.pkl')
for i in range(num_images):
im, _ = self.dataset.pull_image(i)
h, w, _ = im.shape
scale = np.array([[w, h, w, h]])
# preprocess
x = self.transform(im)[0]
x = x.unsqueeze(0).to(self.device)
t0 = time.time()
# forward
bboxes, scores, cls_inds = net(x)
detect_time = time.time() - t0
# rescale
bboxes *= scale
for j in range(len(self.labelmap)):
inds = np.where(cls_inds == j)[0]
if len(inds) == 0:
self.all_boxes[j][i] = np.empty([0, 5], dtype=np.float32)
continue
c_bboxes = bboxes[inds]
c_scores = scores[inds]
c_dets = np.hstack(
(c_bboxes, c_scores[:, np.newaxis])).astype(np.float32,
copy=False)
self.all_boxes[j][i] = c_dets
if i % 500 == 0:
print('im_detect: {:d}/{:d} {:.3f}s'.format(
i + 1, num_images, detect_time))
with open(det_file, 'wb') as f:
pickle.dump(self.all_boxes, f, pickle.HIGHEST_PROTOCOL)
print('Evaluating detections')
self.evaluate_detections(self.all_boxes)
print('Mean AP: ', self.map)
def parse_rec(self, filename):
""" Parse a PASCAL VOC xml file """
tree = ET.parse(filename)
objects = []
for obj in tree.findall('object'):
obj_struct = {}
obj_struct['name'] = obj.find('name').text
obj_struct['pose'] = obj.find('pose').text
obj_struct['truncated'] = int(obj.find('truncated').text)
obj_struct['difficult'] = int(obj.find('difficult').text)
bbox = obj.find('bndbox')
obj_struct['bbox'] = [
int(bbox.find('xmin').text),
int(bbox.find('ymin').text),
int(bbox.find('xmax').text),
int(bbox.find('ymax').text)
]
objects.append(obj_struct)
return objects
def get_output_dir(self, name, phase):
"""Return the directory where experimental artifacts are placed.
If the directory does not exist, it is created.
A canonical path is built using the name from an imdb and a network
(if not None).
"""
filedir = os.path.join(name, phase)
if not os.path.exists(filedir):
os.makedirs(filedir)
return filedir
def get_voc_results_file_template(self, cls):
# VOCdevkit/VOC2007/results/det_test_aeroplane.txt
filename = 'det_' + self.set_type + '_%s.txt' % (cls)
filedir = os.path.join(self.devkit_path, 'results')
if not os.path.exists(filedir):
os.makedirs(filedir)
path = os.path.join(filedir, filename)
return path
def write_voc_results_file(self, all_boxes):
for cls_ind, cls in enumerate(self.labelmap):
if self.display:
print('Writing {:s} VOC results file'.format(cls))
filename = self.get_voc_results_file_template(cls)
with open(filename, 'wt') as f:
for im_ind, index in enumerate(self.dataset.ids):
dets = all_boxes[cls_ind][im_ind]
if dets == []:
continue
# the VOCdevkit expects 1-based indices
for k in range(dets.shape[0]):
f.write(
'{:s} {:.3f} {:.1f} {:.1f} {:.1f} {:.1f}\n'.format(
index[1], dets[k, -1], dets[k, 0] + 1,
dets[k, 1] + 1, dets[k, 2] + 1,
dets[k, 3] + 1))
def do_python_eval(self, use_07=True):
cachedir = os.path.join(self.devkit_path, 'annotations_cache')
aps = []
# The PASCAL VOC metric changed in 2010
use_07_metric = use_07
print('VOC07 metric? ' + ('Yes' if use_07_metric else 'No'))
if not os.path.isdir(self.output_dir):
os.mkdir(self.output_dir)
for i, cls in enumerate(self.labelmap):
filename = self.get_voc_results_file_template(cls)
rec, prec, ap = self.voc_eval(detpath=filename,
classname=cls,
cachedir=cachedir,
ovthresh=0.5,
use_07_metric=use_07_metric)
aps += [ap]
print('AP for {} = {:.4f}'.format(cls, ap))
with open(os.path.join(self.output_dir, cls + '_pr.pkl'),
'wb') as f:
pickle.dump({'rec': rec, 'prec': prec, 'ap': ap}, f)
if self.display:
self.map = np.mean(aps)
print('Mean AP = {:.4f}'.format(np.mean(aps)))
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(
'--------------------------------------------------------------'
)
else:
self.map = np.mean(aps)
print('Mean AP = {:.4f}'.format(np.mean(aps)))
def voc_ap(self, rec, prec, use_07_metric=True):
""" ap = voc_ap(rec, prec, [use_07_metric])
Compute VOC AP given precision and recall.
If use_07_metric is true, uses the
VOC 07 11 point method (default:True).
"""
if use_07_metric:
# 11 point metric
ap = 0.
for t in np.arange(0., 1.1, 0.1):
if np.sum(rec >= t) == 0:
p = 0
else:
p = np.max(prec[rec >= t])
ap = ap + p / 11.
else:
# correct AP calculation
# first append sentinel values at the end
mrec = np.concatenate(([0.], rec, [1.]))
mpre = np.concatenate(([0.], prec, [0.]))
# compute the precision envelope
for i in range(mpre.size - 1, 0, -1):
mpre[i - 1] = np.maximum(mpre[i - 1], mpre[i])
# to calculate area under PR curve, look for points
# where X axis (recall) changes value
i = np.where(mrec[1:] != mrec[:-1])[0]
# and sum (\Delta recall) * prec
ap = np.sum((mrec[i + 1] - mrec[i]) * mpre[i + 1])
return ap
def voc_eval(self,
detpath,
classname,
cachedir,
ovthresh=0.5,
use_07_metric=True):
if not os.path.isdir(cachedir):
os.mkdir(cachedir)
cachefile = os.path.join(cachedir, 'annots.pkl')
# read list of images
with open(self.imgsetpath, 'r') as f:
lines = f.readlines()
imagenames = [x.strip() for x in lines]
if not os.path.isfile(cachefile):
# load annots
recs = {}
for i, imagename in enumerate(imagenames):
recs[imagename] = self.parse_rec(self.annopath % (imagename))
if i % 100 == 0 and self.display:
print('Reading annotation for {:d}/{:d}'.format(
i + 1, len(imagenames)))
# save
if self.display:
print('Saving cached annotations to {:s}'.format(cachefile))
with open(cachefile, 'wb') as f:
pickle.dump(recs, f)
else:
# load
with open(cachefile, 'rb') as f:
recs = pickle.load(f)
# extract gt objects for this class
class_recs = {}
npos = 0
for imagename in imagenames:
R = [obj for obj in recs[imagename] if obj['name'] == classname]
bbox = np.array([x['bbox'] for x in R])
difficult = np.array([x['difficult'] for x in R]).astype(np.bool)
det = [False] * len(R)
npos = npos + sum(~difficult)
class_recs[imagename] = {
'bbox': bbox,
'difficult': difficult,
'det': det
}
# read dets
detfile = detpath.format(classname)
with open(detfile, 'r') as f:
lines = f.readlines()
if any(lines) == 1:
splitlines = [x.strip().split(' ') for x in lines]
image_ids = [x[0] for x in splitlines]
confidence = np.array([float(x[1]) for x in splitlines])
BB = np.array([[float(z) for z in x[2:]] for x in splitlines])
# sort by confidence
sorted_ind = np.argsort(-confidence)
sorted_scores = np.sort(-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, 0.)
ih = np.maximum(iymax - iymin, 0.)
inters = iw * ih
uni = ((bb[2] - bb[0]) * (bb[3] - bb[1]) +
(BBGT[:, 2] - BBGT[:, 0]) *
(BBGT[:, 3] - BBGT[:, 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 / 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 = self.voc_ap(rec, prec, use_07_metric)
else:
rec = -1.
prec = -1.
ap = -1.
return rec, prec, ap
def evaluate_detections(self, box_list):
self.write_voc_results_file(box_list)
self.do_python_eval()
# cuda
if args.cuda:
print('use cuda')
cudnn.benchmark = True
device = torch.device("cuda")
else:
device = torch.device("cpu")
# dataset and evaluator
if args.dataset == 'voc':
data_dir = os.path.join(args.root, 'VOCdevkit')
class_names = VOC_CLASSES
class_indexs = None
num_classes = 20
anchor_size = config.MULTI_ANCHOR_SIZE
dataset = VOCDetection(data_dir=data_dir, img_size=args.img_size)
elif args.dataset == 'coco':
data_dir = os.path.join(args.root, 'COCO')
class_names = coco_class_labels
class_indexs = coco_class_index
num_classes = 80
anchor_size = config.MULTI_ANCHOR_SIZE_COCO
dataset = COCODataset(data_dir=data_dir,
img_size=args.img_size,
image_set='val2017')
else:
print('unknow dataset !! Only support voc and coco !!')
exit(0)
print('use cuda')
cudnn.benchmark = True
device = torch.device("cuda")
else:
device = torch.device("cpu")
# input size
input_size = args.input_size
# dataset and evaluator
if args.dataset == 'voc':
data_dir = os.path.join(args.root, 'VOCdevkit')
class_names = VOC_CLASSES
class_indexs = None
num_classes = 20
dataset = VOCDetection(data_dir=data_dir,
image_sets=[('2007', 'test')])
elif args.dataset == 'coco':
data_dir = os.path.join(args.root, 'COCO')
class_names = coco_class_labels
class_indexs = coco_class_index
num_classes = 80
dataset = COCODataset(data_dir=data_dir, image_set='val')
class_colors = [(np.random.randint(255), np.random.randint(255),
np.random.randint(255)) for _ in range(num_classes)]
# model
model_name = args.version
print('Model: ', model_name)
#Validation data loader
datase_v = TransformedCocoDataset(
cocoRoot + "val2017/val2017", cocoRoot +
"annotations_trainval2017/annotations/instances_val2017.json")
dataLoader_v = DataLoader(datase_v,
batch_size,
num_workers=num_workers,
shuffle=False,
collate_fn=collate_fn,
pin_memory=cuda)
elif dataset == "VOC":
voc_root = "/home/renatkhiz/data/VOCdevkit/"
train_sets = [('2007', 'trainval'), ('2012', 'trainval')]
datase_t = VOCDetection(
voc_root, train_sets,
SSDAugmentation(size=(height, width), mean=(0, 0, 0)),
AnnotationTransform())
dataLoader_t = DataLoader(datase_t,
batch_size,
num_workers=num_workers,
shuffle=True,
collate_fn=collate_fn,
pin_memory=cuda)
model_cpu = ssd.VGGSSD(num_classes=num_classes)
model_init(model_cpu)
if cuda:
model = torch.nn.DataParallel(model_cpu).cuda()
else:
model = model_cpu