def get_FPS(self, image, test_interval):
image_shape = np.array(np.shape(image)[0:2])
#---------------------------------------------------------#
# 给图像增加灰条，实现不失真的resize
#---------------------------------------------------------#
crop_img = letterbox_image(image, [self.input_shape[0],self.input_shape[1]])
#----------------------------------------------------------------------------------#
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
#----------------------------------------------------------------------------------#
photo = np.array(crop_img,dtype = np.float32)[:,:,::-1]
#-----------------------------------------------------------#
# 图片预处理，归一化。获得的photo的shape为[1, 512, 512, 3]
#-----------------------------------------------------------#
photo = np.reshape(preprocess_image(photo), [1, self.input_shape[0], self.input_shape[1], self.input_shape[2]])
preds = self.centernet.predict(photo)
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0])>0:
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[0][:, 1], preds[0][:, 2], preds[0][:, 3]
top_indices = [i for i, conf in enumerate(det_conf) if conf >= self.confidence]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(det_xmin[top_indices],-1),np.expand_dims(det_ymin[top_indices],-1),np.expand_dims(det_xmax[top_indices],-1),np.expand_dims(det_ymax[top_indices],-1)
boxes = centernet_correct_boxes(top_ymin,top_xmin,top_ymax,top_xmax,np.array([self.input_shape[0],self.input_shape[1]]),image_shape)
t1 = time.time()
for _ in range(test_interval):
preds = self.centernet.predict(photo)
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0])>0:
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[0][:, 1], preds[0][:, 2], preds[0][:, 3]
top_indices = [i for i, conf in enumerate(det_conf) if conf >= self.confidence]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(det_xmin[top_indices],-1),np.expand_dims(det_ymin[top_indices],-1),np.expand_dims(det_xmax[top_indices],-1),np.expand_dims(det_ymax[top_indices],-1)
boxes = centernet_correct_boxes(top_ymin,top_xmin,top_ymax,top_xmax,np.array([self.input_shape[0],self.input_shape[1]]),image_shape)
t2 = time.time()
tact_time = (t2 - t1) / test_interval
return tact_time
def detect_image(self, image_id, image):
f = open("./input/detection-results/" + image_id + ".txt", "w")
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
crop_img = letterbox_image(image,
[self.input_shape[0], self.input_shape[1]])
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
# 图片预处理，归一化
photo = np.reshape(
preprocess_image(photo),
[1, self.input_shape[0], self.input_shape[1], self.input_shape[2]])
preds = self.centernet.predict(photo)
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0]) <= 0:
return image
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
# 筛选出其中得分高于confidence的框
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[
0][:, 1], preds[0][:, 2], preds[0][:, 3]
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
# 去掉灰条
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.input_shape[0], self.input_shape[1]]), image_shape)
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = str(top_conf[i])
top, left, bottom, right = boxes[i]
f.write("%s %s %s %s %s %s\n" %
(predicted_class, score[:6], str(int(left)), str(
int(top)), str(int(right)), str(int(bottom))))
f.close()
return
def get_FPS(self, image, test_interval):
image_shape = np.array(np.shape(image)[0:2])
#---------------------------------------------------------#
# 给图像增加灰条，实现不失真的resize
#---------------------------------------------------------#
crop_img = letterbox_image(image,
[self.image_size[0], self.image_size[1]])
#----------------------------------------------------------------------------------#
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
#----------------------------------------------------------------------------------#
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
#-----------------------------------------------------------#
# 图片预处理，归一化。获得的photo的shape为[1, 512, 512, 3]
#-----------------------------------------------------------#
photo = np.reshape(
np.transpose(preprocess_image(photo), (2, 0, 1)),
[1, self.image_size[2], self.image_size[0], self.image_size[1]])
with torch.no_grad():
images = Variable(
torch.from_numpy(np.asarray(photo)).type(torch.FloatTensor))
if self.cuda:
images = images.cuda()
outputs = self.centernet(images)
if self.backbone == 'hourglass':
outputs = [
outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"],
outputs[-1]["reg"]
]
outputs = decode_bbox(outputs[0], outputs[1], outputs[2],
self.image_size, self.confidence, self.cuda)
try:
if self.nms:
outputs = np.array(nms(outputs, self.nms_threhold))
output = outputs[0]
if len(output) > 0:
batch_boxes, det_conf, det_label = output[:, :
4], output[:,
4], output[:,
5]
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:,
0], batch_boxes[:,
1], batch_boxes[:,
2], batch_boxes[:,
3]
top_indices = [
i for i, conf in enumerate(det_conf)
if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices], -1), np.expand_dims(
det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.image_size[0], self.image_size[1]]),
image_shape)
except:
pass
t1 = time.time()
for _ in range(test_interval):
with torch.no_grad():
outputs = self.centernet(images)
if self.backbone == 'hourglass':
outputs = [
outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"],
outputs[-1]["reg"]
]
outputs = decode_bbox(outputs[0], outputs[1], outputs[2],
self.image_size, self.confidence,
self.cuda)
try:
if self.nms:
outputs = np.array(nms(outputs, self.nms_threhold))
output = outputs[0]
if len(output) > 0:
batch_boxes, det_conf, det_label = output[:, :
4], output[:,
4], output[:,
5]
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:,
0], batch_boxes[:,
1], batch_boxes[:,
2], batch_boxes[:,
3]
top_indices = [
i for i, conf in enumerate(det_conf)
if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices], -1), np.expand_dims(
det_ymin[top_indices],
-1), np.expand_dims(det_xmax[top_indices],
-1), np.expand_dims(
det_ymax[top_indices],
-1)
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.image_size[0], self.image_size[1]]),
image_shape)
except:
pass
t2 = time.time()
tact_time = (t2 - t1) / test_interval
return tact_time
def detect_image(self, image):
image_shape = np.array(np.shape(image)[0:2])
#---------------------------------------------------------#
# 给图像增加灰条，实现不失真的resize
#---------------------------------------------------------#
crop_img = letterbox_image(image,
[self.input_shape[0], self.input_shape[1]])
#----------------------------------------------------------------------------------#
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
#----------------------------------------------------------------------------------#
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
#-----------------------------------------------------------#
# 图片预处理，归一化。获得的photo的shape为[1, 512, 512, 3]
#-----------------------------------------------------------#
photo = np.reshape(
preprocess_image(photo),
[1, self.input_shape[0], self.input_shape[1], self.input_shape[2]])
preds = self.get_pred(photo).numpy()
#-------------------------------------------------------#
# 对于centernet网络来讲，确立中心非常重要。
# 对于大目标而言，会存在许多的局部信息。
# 此时对于同一个大目标，中心点比较难以确定。
# 使用最大池化的非极大抑制方法无法去除局部框
# 所以我还是写了另外一段对框进行非极大抑制的代码
# 实际测试中，hourglass为主干网络时有无额外的nms相差不大，resnet相差较大。
#-------------------------------------------------------#
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0]) <= 0:
return image
#-----------------------------------------------------------#
# 将预测结果转换成小数的形式
#-----------------------------------------------------------#
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[
0][:, 1], preds[0][:, 2], preds[0][:, 3]
#-----------------------------------------------------------#
# 筛选出其中得分高于confidence的框
#-----------------------------------------------------------#
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
#-----------------------------------------------------------#
# 去掉灰条部分
#-----------------------------------------------------------#
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.input_shape[0], self.input_shape[1]]), image_shape)
font = ImageFont.truetype(font='model_data/simhei.ttf',
size=np.floor(3e-2 * np.shape(image)[1] +
0.5).astype('int32'))
thickness = max(
(np.shape(image)[0] + np.shape(image)[1]) // self.input_shape[0],
1)
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = top_conf[i]
top, left, bottom, right = boxes[i]
top = top - 5
left = left - 5
bottom = bottom + 5
right = right + 5
top = max(0, np.floor(top + 0.5).astype('int32'))
left = max(0, np.floor(left + 0.5).astype('int32'))
bottom = min(
np.shape(image)[0],
np.floor(bottom + 0.5).astype('int32'))
right = min(
np.shape(image)[1],
np.floor(right + 0.5).astype('int32'))
# 画框框
label = '{} {:.2f}'.format(predicted_class, score)
draw = ImageDraw.Draw(image)
label_size = draw.textsize(label, font)
label = label.encode('utf-8')
print(label, top, left, bottom, right)
if top - label_size[1] >= 0:
text_origin = np.array([left, top - label_size[1]])
else:
text_origin = np.array([left, top + 1])
for i in range(thickness):
draw.rectangle([left + i, top + i, right - i, bottom - i],
outline=self.colors[int(c)])
draw.rectangle(
[tuple(text_origin),
tuple(text_origin + label_size)],
fill=self.colors[int(c)])
draw.text(text_origin,
str(label, 'UTF-8'),
fill=(0, 0, 0),
font=font)
del draw
return image
def detect_image(self, image_id, image):
f = open("./input/detection-results/" + image_id + ".txt", "w")
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
#---------------------------------------------------------#
# 给图像增加灰条，实现不失真的resize
#---------------------------------------------------------#
crop_img = letterbox_image(image,
[self.input_shape[0], self.input_shape[1]])
#----------------------------------------------------------------------------------#
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
#----------------------------------------------------------------------------------#
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
#-----------------------------------------------------------#
# 图片预处理，归一化。获得的photo的shape为[1, 512, 512, 3]
#-----------------------------------------------------------#
photo = np.reshape(
preprocess_image(photo),
[1, self.input_shape[0], self.input_shape[1], self.input_shape[2]])
preds = self.centernet.predict(photo)
#--------------------------------------------------------------------------#
# 对于centernet网络来讲，确立中心非常重要。
# 对于大目标而言，会存在许多的局部信息。
# 此时对于同一个大目标，中心点比较难以确定。
# 使用最大池化的非极大抑制方法无法去除局部框
# 所以我还是写了另外一段对框进行非极大抑制的代码
# 实际测试中，hourglass为主干网络时有无额外的nms相差不大，resnet相差较大。
#---------------------------------------------------------------------------#
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0]) <= 0:
return
#-----------------------------------------------------------#
# 将预测结果转换成小数的形式
#-----------------------------------------------------------#
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
#-----------------------------------------------------------#
# 筛选出其中得分高于confidence的框
#-----------------------------------------------------------#
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[
0][:, 1], preds[0][:, 2], preds[0][:, 3]
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
#-----------------------------------------------------------#
# 去掉灰条部分
#-----------------------------------------------------------#
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.input_shape[0], self.input_shape[1]]), image_shape)
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = str(top_conf[i])
top, left, bottom, right = boxes[i]
f.write("%s %s %s %s %s %s\n" %
(predicted_class, score[:6], str(int(left)), str(
int(top)), str(int(right)), str(int(bottom))))
f.close()
return
def detect_image(self, image):
image_shape = np.array(np.shape(image)[0:2])
crop_img = letterbox_image(image,
[self.image_size[0], self.image_size[1]])
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
# 图片预处理，归一化
photo = np.reshape(
np.transpose(preprocess_image(photo), (2, 0, 1)),
[1, self.image_size[2], self.image_size[0], self.image_size[1]])
with torch.no_grad():
photo = np.asarray(photo)
images = Variable(torch.from_numpy(photo).type(torch.FloatTensor))
if self.cuda:
images = images.cuda()
outputs = self.centernet(images)
if self.backbone == 'hourglass':
outputs = [
outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"],
outputs[-1]["reg"]
]
outputs = decode_bbox(outputs[0], outputs[1], outputs[2],
self.image_size, self.confidence, self.cuda)
#-------------------------------------------------------#
# 对于centernet网络来讲，确立中心非常重要。
# 对于大目标而言，会存在许多的局部信息。
# 此时对于同一个大目标，中心点比较难以确定。
# 使用最大池化的非极大抑制方法无法去除局部框
# 所以我还是写了另外一段对框进行非极大抑制的代码
# 实际测试中，hourglass为主干网络时有无额外的nms相差不大，resnet相差较大。
#-------------------------------------------------------#
try:
if self.nms:
outputs = np.array(nms(outputs, self.nms_threhold))
except:
pass
output = outputs[0]
if len(output) <= 0:
return image
batch_boxes, det_conf, det_label = output[:, :4], output[:,
4], output[:,
5]
# 筛选出其中得分高于confidence的框
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:,
0], batch_boxes[:,
1], batch_boxes[:,
2], batch_boxes[:,
3]
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
# 去掉灰条
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.image_size[0], self.image_size[1]]), image_shape)
font = ImageFont.truetype(font='model_data/simhei.ttf',
size=np.floor(3e-2 * np.shape(image)[1] +
0.5).astype('int32'))
thickness = (np.shape(image)[0] +
np.shape(image)[1]) // self.image_size[0]
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = top_conf[i]
top, left, bottom, right = boxes[i]
top = top - 5
left = left - 5
bottom = bottom + 5
right = right + 5
top = max(0, np.floor(top + 0.5).astype('int32'))
left = max(0, np.floor(left + 0.5).astype('int32'))
bottom = min(
np.shape(image)[0],
np.floor(bottom + 0.5).astype('int32'))
right = min(
np.shape(image)[1],
np.floor(right + 0.5).astype('int32'))
# 画框框
label = '{} {:.2f}'.format(predicted_class, score)
draw = ImageDraw.Draw(image)
label_size = draw.textsize(label, font)
label = label.encode('utf-8')
print(label)
if top - label_size[1] >= 0:
text_origin = np.array([left, top - label_size[1]])
else:
text_origin = np.array([left, top + 1])
for i in range(thickness):
draw.rectangle([left + i, top + i, right - i, bottom - i],
outline=self.colors[int(c)])
draw.rectangle(
[tuple(text_origin),
tuple(text_origin + label_size)],
fill=self.colors[int(c)])
draw.text(text_origin,
str(label, 'UTF-8'),
fill=(0, 0, 0),
font=font)
del draw
return image
def detect_image(self, image_id, image):
f = open("./input/detection-results/" + image_id + ".txt", "w")
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
crop_img = letterbox_image(image,
[self.image_size[0], self.image_size[1]])
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
# 图片预处理，归一化
photo = np.reshape(
np.transpose(preprocess_image(photo), (2, 0, 1)),
[1, self.image_size[2], self.image_size[0], self.image_size[1]])
with torch.no_grad():
photo = np.asarray(photo)
images = Variable(torch.from_numpy(photo).type(torch.FloatTensor))
if self.cuda:
images = images.cuda()
outputs = self.centernet(images)
if self.backbone == 'hourglass':
outputs = [
outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"],
outputs[-1]["reg"]
]
outputs = decode_bbox(outputs[0], outputs[1], outputs[2],
self.image_size, self.confidence, self.cuda)
try:
if self.nms:
outputs = np.array(nms(outputs, self.nms_threhold))
except:
pass
output = outputs[0]
if len(output) <= 0:
return image
batch_boxes, det_conf, det_label = output[:, :4], output[:,
4], output[:,
5]
# 筛选出其中得分高于confidence的框
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:,
0], batch_boxes[:,
1], batch_boxes[:,
2], batch_boxes[:,
3]
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
# 去掉灰条
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.image_size[0], self.image_size[1]]), image_shape)
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = str(top_conf[i])
top, left, bottom, right = boxes[i]
f.write("%s %s %s %s %s %s\n" %
(predicted_class, score[:6], str(int(left)), str(
int(top)), str(int(right)), str(int(bottom))))
f.close()
return
def detect_image(self, image_id, image):
f = open("./input/detection-results/" + image_id + ".txt", "w")
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
#---------------------------------------------------------#
# 给图像增加灰条，实现不失真的resize
#---------------------------------------------------------#
crop_img = letterbox_image(image,
[self.image_size[0], self.image_size[1]])
#----------------------------------------------------------------------------------#
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
#----------------------------------------------------------------------------------#
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
#-----------------------------------------------------------#
# 图片预处理，归一化。获得的photo的shape为[1, 512, 512, 3]
#-----------------------------------------------------------#
photo = np.reshape(
np.transpose(preprocess_image(photo), (2, 0, 1)),
[1, self.image_size[2], self.image_size[0], self.image_size[1]])
with torch.no_grad():
images = Variable(
torch.from_numpy(np.asarray(photo)).type(torch.FloatTensor))
if self.cuda:
images = images.cuda()
outputs = self.centernet(images)
if self.backbone == 'hourglass':
outputs = [
outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"],
outputs[-1]["reg"]
]
#-----------------------------------------------------------#
# 利用预测结果进行解码
#-----------------------------------------------------------#
outputs = decode_bbox(outputs[0], outputs[1], outputs[2],
self.image_size, self.confidence, self.cuda)
#-------------------------------------------------------#
# 对于centernet网络来讲，确立中心非常重要。
# 对于大目标而言，会存在许多的局部信息。
# 此时对于同一个大目标，中心点比较难以确定。
# 使用最大池化的非极大抑制方法无法去除局部框
# 所以我还是写了另外一段对框进行非极大抑制的代码
# 实际测试中，hourglass为主干网络时有无额外的nms相差不大，resnet相差较大。
#-------------------------------------------------------#
try:
if self.nms:
outputs = np.array(nms(outputs, self.nms_threhold))
except:
pass
output = outputs[0]
if len(output) <= 0:
return image
batch_boxes, det_conf, det_label = output[:, :
4], output[:,
4], output[:,
5]
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:,
0], batch_boxes[:,
1], batch_boxes[:,
2], batch_boxes[:,
3]
#-----------------------------------------------------------#
# 筛选出其中得分高于confidence的框
#-----------------------------------------------------------#
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
#-----------------------------------------------------------#
# 去掉灰条部分
#-----------------------------------------------------------#
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.image_size[0], self.image_size[1]]),
image_shape)
for i, c in enumerate(top_label_indices):
predicted_class = self.class_names[int(c)]
score = str(top_conf[i])
top, left, bottom, right = boxes[i]
f.write("%s %s %s %s %s %s\n" %
(predicted_class, score[:6], str(int(left)), str(
int(top)), str(int(right)), str(int(bottom))))
f.close()
return
def detect_image(self, image_id, image, results):
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
crop_img = letterbox_image(image,
[self.input_shape[0], self.input_shape[1]])
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
photo = np.array(crop_img, dtype=np.float32)[:, :, ::-1]
# 图片预处理，归一化
photo = np.reshape(
preprocess_image(photo),
[1, self.input_shape[0], self.input_shape[1], self.input_shape[2]])
preds = self.centernet.predict(photo)
if self.nms:
preds = np.array(nms(preds, self.nms_threhold))
if len(preds[0]) <= 0:
return results
preds[0][:, 0:4] = preds[0][:, 0:4] / (self.input_shape[0] / 4)
# 筛选出其中得分高于confidence的框
det_label = preds[0][:, -1]
det_conf = preds[0][:, -2]
det_xmin, det_ymin, det_xmax, det_ymax = preds[0][:, 0], preds[
0][:, 1], preds[0][:, 2], preds[0][:, 3]
top_indices = [
i for i, conf in enumerate(det_conf) if conf >= self.confidence
]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(
det_xmin[top_indices],
-1), np.expand_dims(det_ymin[top_indices], -1), np.expand_dims(
det_xmax[top_indices],
-1), np.expand_dims(det_ymax[top_indices], -1)
# 去掉灰条
boxes = centernet_correct_boxes(
top_ymin, top_xmin, top_ymax, top_xmax,
np.array([self.input_shape[0], self.input_shape[1]]), image_shape)
for i, c in enumerate(top_label_indices):
result = {}
predicted_class = self.class_names[int(c)]
top, left, bottom, right = boxes[i]
top = max(0, np.floor(top + 0.5).astype('int32'))
left = max(0, np.floor(left + 0.5).astype('int32'))
bottom = min(image.size[1], np.floor(bottom + 0.5).astype('int32'))
right = min(image.size[0], np.floor(right + 0.5).astype('int32'))
result["image_id"] = int(image_id)
result["category_id"] = clsid2catid[c]
result["bbox"] = [
float(left),
float(top),
float(right - left),
float(bottom - top)
]
result["score"] = float(top_conf[i])
results.append(result)
return results
def detect_image(self, image_id, image, results):
self.confidence = 0.01
self.nms_threhold = 0.5
image_shape = np.array(np.shape(image)[0:2])
crop_img = letterbox_image(image, [self.image_size[0],self.image_size[1]])
# 将RGB转化成BGR，这是因为原始的centernet_hourglass权值是使用BGR通道的图片训练的
photo = np.array(crop_img,dtype = np.float32)[:,:,::-1]
# 图片预处理，归一化
photo = np.reshape(np.transpose(preprocess_image(photo), (2, 0, 1)),[1,self.image_size[2],self.image_size[0],self.image_size[1]])
with torch.no_grad():
photo = np.asarray(photo)
images = Variable(torch.from_numpy(photo).type(torch.FloatTensor))
if self.cuda:
images = images.cuda()
outputs = self.centernet(images)
if self.backbone=='hourglass':
outputs = [outputs[-1]["hm"].sigmoid(), outputs[-1]["wh"], outputs[-1]["reg"]]
outputs = decode_bbox(outputs[0],outputs[1],outputs[2],self.image_size,self.confidence,self.cuda)
try:
if self.nms:
outputs = np.array(nms(outputs,self.nms_threhold))
except:
pass
output = outputs[0]
if len(output)<=0:
return image
batch_boxes, det_conf, det_label = output[:,:4], output[:,4], output[:,5]
# 筛选出其中得分高于confidence的框
det_xmin, det_ymin, det_xmax, det_ymax = batch_boxes[:, 0], batch_boxes[:, 1], batch_boxes[:, 2], batch_boxes[:, 3]
top_indices = [i for i, conf in enumerate(det_conf) if conf >= self.confidence]
top_conf = det_conf[top_indices]
top_label_indices = det_label[top_indices].tolist()
top_xmin, top_ymin, top_xmax, top_ymax = np.expand_dims(det_xmin[top_indices],-1),np.expand_dims(det_ymin[top_indices],-1),np.expand_dims(det_xmax[top_indices],-1),np.expand_dims(det_ymax[top_indices],-1)
# 去掉灰条
boxes = centernet_correct_boxes(top_ymin,top_xmin,top_ymax,top_xmax,np.array([self.image_size[0],self.image_size[1]]),image_shape)
for i, c in enumerate(top_label_indices):
result = {}
predicted_class = self.class_names[int(c)]
top, left, bottom, right = boxes[i]
top = max(0, np.floor(top + 0.5).astype('int32'))
left = max(0, np.floor(left + 0.5).astype('int32'))
bottom = min(image.size[1], np.floor(bottom + 0.5).astype('int32'))
right = min(image.size[0], np.floor(right + 0.5).astype('int32'))
result["image_id"] = int(image_id)
result["category_id"] = clsid2catid[c]
result["bbox"] = [float(left),float(top),float(right-left),float(bottom-top)]
result["score"] = float(top_conf[i])
results.append(result)
return results