def prep_frame(ftpts, frame, camera, errors, occupants, ped_bboxes, veh_bboxes,
classes):
pix_real = camera["im-gps"]
real_pix = camera["gps-im"]
origin = camera["estimated_camera_location"]
frame_size = camera["frame_size"]
print(frame_size)
#frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if len(ped_bboxes) > 0:
frame = utils.draw_bbox(frame,
ped_bboxes,
classes,
show_label=True,
redact=True)
frame, x = pr.draw_radius(frame, ftpts, real_pix, pix_real, origin)
utils.overlay_occupancy(frame, errors, occupants, frame_size)
if len(veh_bboxes) > 0:
frame = utils.draw_bbox(frame,
veh_bboxes,
classes,
show_label=False,
redact=False)
#result = cv2.cvtColor(frame, cv2.COLOR_RGB2BGR)
return frame
def test_predict():
from sklearn.model_selection import train_test_split
import skimage.io as io
with open('bboxGen/cropping.txt', 'rt') as f:
df = f.read().split('\n')[:-1]
df = [line.split(',') for line in df]
coords = []
for p, *coord in df:
coords.append((p, [(int(coord[i]), int(coord[i+1]))
for i in range(0, len(coord), 2)]))
trn_fnms, val_fnms = train_test_split(coords, test_size=200, random_state=42)
model = bbox_model()
model.load_state_dict(torch.load('bboxGen/logs/model-74.pth'))
model.to(device)
model.eval()
imgnm, gtbox = val_fnms[np.random.randint(0, high=199)]
gtbox = bounding_rectangle(gtbox, 9999999)
with torch.no_grad():
pred = predict(model, 'rawdata/bboxes/'+imgnm)
image = io.imread('rawdata/bboxes/'+imgnm)
draw_bbox(image, pred, gtbox)
print(gtbox)
print(pred)
fnms = os.listdir('rawdata/train')
trn_img = os.path.join('rawdata/train', fnms[np.random.randint(0, len(fnms))])
with torch.no_grad():
pred = predict(model, trn_img)
image = io.imread(trn_img)
draw_bbox(image, pred)
def test_camera():
""" Simple test srcipt; requires /dev/video0 """
app = ObjectDetectorTF(gpu=False, cmap={1: 'person'})
cam = cv2.VideoCapture(0)
fps = []
while True:
ret, img = cam.read()
if not ret:
print('camera capture failed')
break
t0 = time.time()
res = app(img)
t1 = time.time()
fps.append(1.0 / (t1 - t0 + 1e-9))
msk = (res['score'] > 0.5)
cls = res['class'][msk]
box = res['box'][msk]
score = res['score'][msk]
for box_, cls_ in zip(box, cls):
#ry0,rx0,ry1,rx1 = box_ # relative
draw_bbox(img, box_, str(cls_))
cv2.imshow('win', img)
k = cv2.waitKey(1)
if k in [ord('q'), 27]:
print('quitting...')
break
print('average fps: {}'.format(np.mean(fps[-100:])))
def run(args):
model = build_model()
(clf, class_names) = read_classifier(
os.path.join(args.data_path, 'classifier.pickle'))
# if classifier is none we only have one face
if clf is None:
verified_embedding, only_class = read_only_embedding(args.data_path)
cap = cv2.VideoCapture(0)
if (cap.isOpened() == False):
print("Error opening video stream or file")
face_detector = FaceDetector()
while (cap.isOpened()):
# Capture frame-by-frame
ret, frame = cap.read()
if ret == True:
# Detect image and write it
faces = face_detector.detect_faces(frame)
for face in faces:
x, y, w, h = face
cropped = frame[y:y + h, x:x + w]
cropped = cv2.resize(cropped, (96, 96))
cropped = np.around(convert_image(cropped), decimals=12)
embedding = model.predict(np.array([cropped]))
if clf is None:
dist = np.linalg.norm(verified_embedding - embedding)
match = dist < 0.7
label = only_class if match else "Unknown"
if args.debug:
label += ' (d: {})'.format(round(dist, 2))
else:
predictions = clf.predict_proba(embedding)
pred_class = np.argmax(predictions, axis=1)[0]
score = round(np.max(predictions) * 100, 2)
match = score > 70
name = class_names[pred_class]
label = '{} ({}%)'.format(name, score)
color = (0, 255, 0) if match else (0, 0, 255)
draw_bbox(frame, x, y, x + w, y + h, label=label, color=color)
cv2.imshow('Frame', frame)
# Press Q on keyboard to exit
if cv2.waitKey(25) & 0xFF == ord('q'):
break
else:
break
cap.release()
cv2.destroyAllWindows()
def capture(named_path, data_path, count):
cap = cv2.VideoCapture(0)
# Check if camera opened successfully
if (cap.isOpened() == False):
print("Error opening video stream or file")
captured_counter = 0
face_detector = FaceDetector()
model = build_model()
while (cap.isOpened() and captured_counter < count):
# Capture frame-by-frame
ret, frame = cap.read()
if ret == True:
# Show progress bar
draw_progressbar(frame, (captured_counter / count))
# Detect image and write it
faces = face_detector.detect_faces(frame)
if len(faces) > 0:
# Per person path
file_path = os.path.join(named_path,
str(captured_counter + 1) + '.jpg')
print('Writing capture: ' + file_path)
face = faces[0] # Assume it's the only face
x, y, w, h = face
cropped = frame[y:y + h, x:x + w]
cropped = cv2.resize(cropped, (96, 96))
cv2.imwrite(file_path, cropped)
captured_counter += 1
draw_bbox(frame, x, y, x + w, y + h, label="Face detected")
cv2.imshow('Frame', frame)
# Press Q on keyboard to exit
if cv2.waitKey(25) & 0xFF == ord('q'):
break
# Break the loop
else:
break
# When everything done, release the video capture object
cap.release()
cv2.destroyAllWindows()
# Build and Write the embedding file for this person
build_embedding(model, named_path)
# Rebuild the classifier
build_classifier(data_path)
print('Done!')
def detect_video(predict_func, video_path, class_name, input_size, output_path=""):
# read video
vid = cv.VideoCapture(video_path)
video_size = (int(vid.get(cv.CAP_PROP_FRAME_WIDTH)), int(vid.get(cv.CAP_PROP_FRAME_HEIGHT)))
if not vid.isOpened():
raise IOError("Couldn't open webcam or video")
video_FourCC = int(vid.get(cv.CAP_PROP_FOURCC))
video_fps = vid.get(cv.CAP_PROP_FPS)
isOutput = True if output_path != "" else False
if isOutput:
out = cv.VideoWriter(output_path, video_FourCC, video_fps, video_size)
# fps info
accum_time = 0
curr_fps = 0
fps = "FPS: ??"
prev_time = timer()
# generate color map
colors = gen_colors(len(class_name))
# read frames
while True:
return_value, org_frame = vid.read()
if return_value == False:
break
# detect
# frame = squar_crop(frame, cropSize)
frame = cv.resize(org_frame, input_size)
frame = cv.cvtColor(frame, cv.COLOR_BGR2RGB)
boxes, scores, classes = predict_func(frame)
draw_bbox(org_frame, class_name, boxes.numpy(), scores.numpy(), classes.numpy(), colors)
curr_time = timer()
exec_time = curr_time - prev_time
prev_time = curr_time
accum_time = accum_time + exec_time
curr_fps = curr_fps + 1
if accum_time > 1:
accum_time = accum_time - 1
fps = "FPS: " + str(curr_fps)
curr_fps = 0
cv.putText(org_frame, text=fps, org=(3, 15), fontFace=cv.FONT_HERSHEY_SIMPLEX,
fontScale=0.50, color=(0, 255, 0), thickness=2)
cv.namedWindow("result", cv.WINDOW_NORMAL)
cv.imshow("result", org_frame)
if isOutput:
out.write(org_frame)
if cv.waitKey(1) & 0xFF == ord('q'):
break
cv.destroyAllWindows()
def detect_img(predict_func, img_path, class_name, input_size):
orgimg = cv.imread(img_path)
img = cv.resize(orgimg, (416, 416))
img = cv.cvtColor(img, cv.COLOR_BGR2RGB)
colors = gen_colors(len(class_name))
start = timer()
boxes, scores, classes = predict_func(img)
end = timer()
print('spent time: %.3fs' % (end - start))
draw_bbox(orgimg, class_name, boxes.numpy(), scores.numpy(), classes.numpy(), colors)
cv.imshow('img', orgimg)
cv.waitKey()
def predict_img(self,
raw_img,
random_color=True,
plot_img=True,
figsize=(10, 10),
show_text=True,
return_output=False):
print('img shape: ', raw_img.shape)
img = self.preprocess_img(raw_img)
imgs = np.expand_dims(img, axis=0)
pred_output = self.inference_model.predict(imgs)
detections = get_detection_data(img=raw_img,
model_outputs=pred_output,
class_names=self.class_names)
output_img = draw_bbox(raw_img,
detections,
cmap=self.class_color,
random_color=random_color,
figsize=figsize,
show_text=show_text,
show_img=plot_img)
if return_output:
return output_img, detections
else:
return detections
def predict(self):
np.set_printoptions(threshold=np.inf)
image_path = './414162.jpg'
image = np.array(cv2.imread(image_path))
image_shape = image.shape
print("image_shape: ", image_shape)
image = np.copy(image)
image_data = utils.image_preprocess(image,
[self.input_size, self.input_size])
image_data = image_data[np.newaxis, ...]
pred_bbox = self.sess.run([self.pred_bbox],
feed_dict={
self.input: image_data,
self.training: False
})
pred_bbox = np.array(pred_bbox[0])
pred_bbox = utils.postprocess_boxes(pred_bbox, image_shape, 416, 0.5)
print("pred_bbox shape: ", pred_bbox.shape)
pred_bbox = utils.nms(pred_bbox, 0.45)
print("pred_bbox after: ", pred_bbox)
image = utils.draw_bbox(image, pred_bbox, show_label=True)
cv2.imwrite('./test.jpg', image)
def predict(model, detection, img_path, threshold=0.6):
"""
result is coord_list is shape = (number of plates, (x1, y1, x2, y2, prediction_rate))
"""
source_image = cv2.imread(img_path)
img_w, img_h, c = source_image.shape
possible_plates = detection.detect_plate(source_image)
coord_list = []
l = len(possible_plates)
if l > 0:
x_in = np.zeros((l, config.INPUT_HEIGHT, config.INPUT_WIDTH, 3))
count = 0
for count, plate in enumerate(possible_plates):
img_temp = plate.get_plate_img(source_image)
img_temp = reshape_img(img_temp)
x_in[count] = img_temp
pr = model.predict(x_in)
print(pr)
pr = pr[..., -1]
for i in range(len(pr)):
if pr[i] > threshold:
temp_box = list(possible_plates[i].get_extend_box())
temp_box.append(pr[i])
coord_list.append(temp_box)
result_img = source_image
for coord in coord_list:
result_img = draw_bbox(result_img, coord[0:4])
return coord_list, result_img
def is_img(img_cv, color):
j = 0
if len(img_cv) != 0:
print("---1312--------------")
for i in range (len(img_cv)):
im_cv_r = cv2.resize(img_cv[i], (1300, 414))
gray = cv2.cvtColor(im_cv_r, cv2.COLOR_BGR2GRAY)
equ = cv2.equalizeHist(gray)
gaussian = cv2.GaussianBlur(gray, (3, 3), 0, 0, cv2.BORDER_DEFAULT)
median = cv2.medianBlur(gaussian, 3)
original_image = median
original_image_size = original_image.shape[:2]
image_data = utils.image_preporcess(np.copy(original_image), [input_size, input_size])
image_data = image_data[np.newaxis, ...]
data = json.dumps({"signature_name": "serving_default",
"instances": image_data.tolist()})
headers = {"content-type": "application/json"}
num_classes=65
json_response = requests.post(
'http://tf:port/v1/models/yolov3:predict', data=data, headers=headers)
predictions = json.loads(json_response.text)['predictions']
pred_sbbox, pred_mbbox, pred_lbbox =predictions[0]['pred_sbbox'],predictions[0]['pred_mbbox'],predictions[0]['pred_lbbox']
pred_bbox = np.concatenate([np.reshape(pred_sbbox, (-1, 5 + num_classes)),
np.reshape(pred_mbbox, (-1, 5 + num_classes)),
np.reshape(pred_lbbox, (-1, 5 + num_classes))], axis=0)
bboxes = utils.postprocess_boxes(pred_bbox, original_image_size, input_size, 0.3)
bboxes = utils.nms(bboxes, 0.45, method='nms')
if np.array(bboxes).shape[0] > 6:
image = utils.draw_bbox(im_cv_r, bboxes)
# print(image)
name = color +'im' + str(i) + '.jpg'
path = os.path.join("./pre_out/", name)
cv2.imwrite(path,image)
print("-------------")
def main(model_path, img_folder, save_path, gpu_id):
if os.path.exists(save_path):
shutil.rmtree(save_path, ignore_errors=True)
if not os.path.exists(save_path):
os.makedirs(save_path)
save_img_folder = os.path.join(save_path, 'img')
if not os.path.exists(save_img_folder):
os.makedirs(save_img_folder)
save_txt_folder = os.path.join(save_path, 'result')
if not os.path.exists(save_txt_folder):
os.makedirs(save_txt_folder)
img_paths = [os.path.join(img_folder, x) for x in os.listdir(img_folder)]
model = Pytorch_model(model_path, gpu_id=gpu_id)
total_frame = 0.0
total_time = 0.0
for img_path in tqdm(img_paths):
img_name = os.path.basename(img_path).split('.')[0]
save_name = os.path.join(save_txt_folder, 'res_' + img_name + '.txt')
_, boxes_list, t = model.predict(img_path)
total_frame += 1
total_time += t
img = draw_bbox(img_path, boxes_list, color=(0, 0, 255))
cv2.imwrite(os.path.join(save_img_folder, '{}.jpg'.format(img_name)), img)
np.savetxt(save_name, boxes_list.reshape(-1, 8), delimiter=',', fmt='%d')
print('fps:{}'.format(total_frame / total_time))
return save_txt_folder
def predict(data_loader, net, config):
net.eval()
for i, (imgs, reg_targets, cls_targets) in enumerate(data_loader):
data = Variable(imgs.cuda(async=True), volatile=True)
reg_preds, cls_preds = net(data)
encoder = DataEncoder(config)
for j in range(len(imgs)):
img = imgs[j]
reg_pred = reg_preds.cpu()[j]
cls_pred = cls_preds.cpu()[j]
reg_target = reg_targets[j]
cls_target = cls_targets[j]
bboxes, scores = encoder.decode(reg_pred.data, cls_pred.data.squeeze(), [config.img_max_size, config.img_max_size])
bboxes2, scores2 = encoder.decode(reg_target, cls_target, [config.img_max_size, config.img_max_size])
img = np.transpose(img.numpy(), (1, 2, 0))
img = ((img * config.sigma + config.mu) * 255).astype(np.uint8)
draw_bbox(img, bboxes.numpy(), scores.numpy(), '/home/storage/lsy/fashion/predictions/'+config.clothes+'/%d-%d.png' % (i, j), bboxes2.numpy())
def __call__(self, IMG):
IMG_ = pad_img(IMG)
[NMS_IDX, BBOX, TOPK_CLASS, TOPK_SCORE] = self.sess.run([self.nms_idx, self.bbox, self.topK_class, self.topK_score], feed_dict={self.inputs: IMG_[np.newaxis] / 127.5 - 1.0, self.is_training: True})
for i in NMS_IDX:
if TOPK_SCORE[i] > 0.5:
IMG = draw_bbox(IMG, recover_ImgAndBbox_scale(IMG, BBOX[i]), CLASSES[TOPK_CLASS[i]])
# IMG_ = draw_bbox(IMG_, np.int32(BBOX[i]), CLASSES[TOPK_CLASS[i]])
return IMG
def predict(self, frame, resize=True):
frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
if not resize:
oryginal = frame
frame = cv2.resize(frame, (self.input_size, self.input_size))
image_data = frame / 255.
image_data = image_data[np.newaxis, ...].astype(
np.float32
) # Creates a single batch shape: (1, input_size, input_size, 3)
self.interpreter.set_tensor(self.input_details[0]['index'], image_data)
self.interpreter.invoke()
pred = [
self.interpreter.get_tensor(self.output_details[i]['index'])
for i in range(len(self.output_details))
]
boxes, pred_conf = utils.filter_boxes(
pred[0],
pred[1],
score_threshold=0.25,
input_shape=tf.constant([self.input_size, self.input_size]))
boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
scores=tf.reshape(
pred_conf,
(tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
max_output_size_per_class=50,
max_total_size=50,
iou_threshold=self.iou,
score_threshold=self.treshold,
)
pred_bbox = [
boxes.numpy(),
scores.numpy(),
classes.numpy(),
valid_detections.numpy()
]
if not resize:
image = utils.draw_bbox(oryginal, pred_bbox, self.classes)
else:
image = utils.draw_bbox(frame, pred_bbox, self.classes)
result = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
return result
def main(_argv):
config = ConfigProto()
# config.gpu_options.allow_growth = True
session = InteractiveSession(config=config)
# model_module = sh_digit
# if FLAGS.model == 'sh_digit':
# model_module = sh_digit
# STRIDES = model_module.STRIDES
# ANCHORS = model_module.ANCHORS
# NUM_CLASS = model_module.NUM_CLASS
# XYSCALE = model_module.XYSCALE
original_image = cv2.imread(FLAGS.image_path)
original_image = cv2.cvtColor(original_image, cv2.COLOR_BGR2RGB)
image_data = cv2.resize(original_image, tuple(FLAGS.image_size_wh))
image_data = image_data / 255.
images_data = []
for i in range(1):
images_data.append(image_data)
images_data = np.asarray(images_data).astype(np.float32)
saved_model_loaded = tf.saved_model.load(FLAGS.weights, tags=[tag_constants.SERVING])
infer = saved_model_loaded.signatures['serving_default']
batch_data = tf.constant(images_data)
pred_bbox = infer(batch_data)
for key, value in pred_bbox.items():
boxes = value[:, :, 0:4]
pred_conf = value[:, :, 4:]
print(boxes)
boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
scores=tf.reshape(
pred_conf, (tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
max_output_size_per_class=50,
max_total_size=50,
iou_threshold=FLAGS.iou,
score_threshold=FLAGS.score
)
print(boxes.numpy())
pred_bbox = [boxes.numpy(), scores.numpy(), classes.numpy(), valid_detections.numpy()]
image = utils.draw_bbox(original_image, pred_bbox,
classes=utils.read_class_names('cfg/digit/digit.names'))
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
cv2.imshow('result', image)
cv2.waitKey()
def test_image(img='/tmp/image1.jpg'):
"""
Simple test script; requires /tmp/image1.jpg
"""
app = ObjectDetectorTF(model='model')
img = cv2.imread(img)
res = app(img)
msk = (res['score'] > 0.5)
cls = res['class'][msk]
box = res['box'][msk]
score = res['score'][msk]
print('score', score)
for box_, cls_ in zip(box, cls):
#ry0,rx0,ry1,rx1 = box_ # relative
draw_bbox(img, box_, str(cls_))
cv2.imshow('win', img)
cv2.waitKey(0)
def object_detect(input_path="./road.jpg", output_path='./demo.jpg'):
img_size = 608
num_channels = 3
# image_path = "./docs/images/sample_computer.jpg"
image_path = input_path # 调用图片,示例："./docs/images/sample_computer.jpg"
original_image = cv2.imread(image_path)
# original_image = cv2.cvtColor(original_image, cv2.COLOR_BGR2RGB)
image_data = utils.image_preporcess(np.copy(original_image),
[img_size, img_size]) # 图片处理成608*608*3
# print(image_data.shape)
plt.imshow(image_data)
plt.show()
yolov3_api = "http://localhost:8501/v1/models/yolov3:predict" # 刚刚产生的接口
image_data_yolo_list = image_data[np.newaxis, :].tolist() # 转化为多维数组矩阵
headers = {"Content-type": "application/json"}
r = requests.post(yolov3_api,
headers=headers,
data=json.dumps({
"signature_name": "predict",
"instances": image_data_yolo_list
})).json() #post请求
# print('r',r) # 19, 19, 85 = 30685
# {'error': 'Input to reshape is a tensor with 18411 values, but the requested shape requires a multiple of 30685\n\t [[{{node pred_multi_scale/Reshape_2}}]]'}
# 18411 的因子 [3, 17, 19, 51, 57, 323, 361, 969, 1083, 6137]
output = np.array(r['predictions'])
# print(output.shape)
# (63, 19, 19, 85) reduction factor 注：衰减系数以及步长：32 608/32=19 85 = 80类+1可能性+4个坐标
# 416 x 416 则为 13*13
output = np.reshape(
output,
(-1, 85
)) # 这一步处理成 22743*85的维度（63*19*19 =22743， 85 = 80类+1可能性+4个坐标,根据自己数据集改）
# print(output.shape)
original_image_size = original_image.shape[:2]
bboxes = utils.postprocess_boxes(
output, original_image_size, img_size,
0.3) # 这一步是将所有可能的预测信息提取出来，主要是三类：类别，可能性，坐标值。
bboxes = utils.nms(bboxes, 0.45,
method='nms') # 这一步是 将刚刚提取出来的信息进行筛选，返回最好的预测值，同样是三类。
image = utils.draw_bbox(original_image, bboxes) # 这一步是把结果画到新图上面。
image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
image = Image.fromarray(image)
image.show()
image.save(output_path) # 保存图片到本地
def main(filename):
flow.load_variables(flow.checkpoint.get(cfg.TEST.WEIGHT_FILE))
image = cv2.imread(filename)
bboxes = predict(image)
image = utils.draw_bbox(image, bboxes, [
'car',
])
image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image = Image.fromarray(image)
image.show()
def detect_img(model_result, input_size, fp_src, fp_dst):
original_image = cv2.imread(fp_src)
original_image_draw = cv2.cvtColor(original_image, cv2.COLOR_BGR2RGB)
original_image_size = original_image.shape[:2]
image_data = utils.image_preporcess(np.copy(original_image),
[input_size, input_size])
image_data = np.array(image_data).astype(np.float32)
bboxes = model_result.model_reponse(image_data, original_image_size)
image = utils.draw_bbox(original_image_draw, bboxes)
image = Image.fromarray(image)
image.save(fp_dst)
def text_predict(img):
# img = cv2.imread(imgpath)
preds, boxes_list, rects_re, t = text_handle.predict(
img, long_size=pse_long_size)
img2 = draw_bbox(img, boxes_list, color=(0, 255, 0))
cv2.imwrite("debug_im/draw.jpg", img2)
result = crnnRec(np.array(img), rects_re)
return result
def detect_image(self,
image_path=None,
output_path=None,
input_size=416,
show=False,
score_threshold=0.3,
iou_threshold=0.45,
rectangle_colors=''):
if image_path is not None:
original_image = cv2.imread(image_path)
original_image = cv2.cvtColor(original_image, cv2.COLOR_BGR2RGB)
original_image = cv2.cvtColor(original_image, cv2.COLOR_BGR2RGB)
image_data = image_preprocess(np.copy(original_image),
[input_size, input_size])
image_data = tf.expand_dims(image_data, 0)
# it gives output in three different scale
pred_bbox = self.tiny_YoloV3.predict(image_data)
print(pred_bbox[0].shape)
print(pred_bbox[1].shape)
pred_bbox = [
tf.reshape(x, (-1, tf.shape(x)[-1])) for x in pred_bbox
]
pred_bbox = tf.concat(pred_bbox, axis=0)
# print(pred_bbox)
bboxes = postprocess_boxes(pred_bbox, original_image, input_size,
score_threshold)
print(bboxes.shape)
bboxes = nms(bboxes, iou_threshold, method='nms')
print(bboxes[0].shape)
print(len(bboxes))
image = draw_bbox(original_image,
bboxes,
CLASSES=self.CLASSES,
rectangle_colors=rectangle_colors)
# print(image.shape)
if output_path is not None:
cv2.imwrite(output_path, image)
if show:
# Show the image
cv2.imshow("predicted image", image)
# Load and hold the image
cv2.waitKey(0)
# To close the window after the required kill value was provided
cv2.destroyAllWindows()
return image
def predict_nonms(self,
img_path,
iou_threshold=0.413,
score_threshold=0.1):
raw_img = cv2.imread(img_path)
print('img shape: ', raw_img.shape)
img = self.preprocess_img(raw_img)
imgs = np.expand_dims(img, axis=0)
yolov4_output = self.yolo_model.predict(imgs)
output = yolov4_head(yolov4_output, self.num_classes, self.anchors,
self.xyscale)
pred_output = nms(output, self.img_size, self.num_classes,
iou_threshold, score_threshold)
pred_output = [p.numpy() for p in pred_output]
detections = get_detection_data(img=raw_img,
model_outputs=pred_output,
class_names=self.class_names)
draw_bbox(raw_img,
detections,
cmap=self.class_color,
random_color=True)
return detections
def text_predict(img):
# img = cv2.imread(imgpath)
if det_model_type == "pse_mobilenetv2":
preds, boxes_list, rects_re, t = text_handle.predict(
img, long_size=pse_long_size)
else:
boxes_list, score_list = text_handle.process(img)
img2 = draw_bbox(img, boxes_list, color=(0, 255, 0))
cv2.imwrite("debug_im/draw.jpg", img2)
# result = crnnRec(np.array(img), rects_re)
result = crnnRecWithBox(np.array(img), boxes_list)
return result
def yolo_predict(img_input):
sample = source = cv2.imread(img_input)
sample = transform_img(sample, IMAGE_SIZE)
sample_shape = sample.shape
result = np.expand_dims(sample, 0)
result = model.predict(result)
result = decode_output(result, sample_shape, GRID_SIZE)
source_shape = source.shape
result = get_original_bbox(result, source_shape, IMAGE_SIZE)
img = draw_bbox(source, result)
# show_img(img, str(source_shape))
return img
def two(self, path1, path2=None):
lines = os.listdir(path1)
for num, line in enumerate(lines):
annotation = line.split()
# print("annotation109",annotation,type(annotation))
image_path = os.path.join(path1, "".join(annotation))
# print("image_path",image_path)
image = cv2.imread(image_path)
bboxes_pr = self.predict(image)
image = utils.draw_bbox(image,
bboxes_pr,
show_label=self.show_label)
if path2 != None:
cv2.imwrite(path2 + "{}".format(num) + ".jpg", image)
cv2.imshow('t', image)
cv2.waitKey(0)
cv2.destroyAllWindows()
def detect_video(video_filename, show=False, interest_rect=None):
cap = cv2.VideoCapture(video_filename)
assert cap.isOpened()
bboxs = []
while True:
res, img = cap.read()
if not res:
break
start_time = time.time()
detect_img = img
if interest_rect is not None:
detect_img = img.copy()
image_h, image_w, _ = detect_img.shape
l, t, r, b = double_rect(interest_rect)
l = max(0, l)
t = max(0, t)
r = min(r, image_w)
b = min(b, image_h)
detect_img = detect_img[t:b, l:r, ...]
bbox = predict(detect_img)
for box in bbox:
box[0] = box[0] + l
box[2] = box[2] + l
box[1] = box[1] + t
box[3] = box[3] + t
else:
bbox = predict(detect_img)
bboxs.append(bbox)
if show:
img = utils.draw_bbox(img, bbox, [
'car',
])
consume = (time.time() - start_time) * 1000
consume = max(0, consume)
cv2.imshow('video', img)
cv2.waitKey(max(1, 45 - int(consume)))
return bboxs
label_reader = GroundtruthReader(args.groundtruth)
bundled_reader = zip(frame_reader, label_reader)
tracker = None
net = models.vgg16_bn(pretrained=True)
patch_video_writer = None
feature_video_writer = None
for frame, gt_bbox in zip(frame_reader, label_reader):
if tracker is None:
tracker = Tracker(frame, gt_bbox)
continue
bbox = tracker.track(frame)
draw_bbox(frame, bbox, (255, 0, 0))
draw_bbox(frame, gt_bbox, (0, 255, 0))
cv2.imshow("Demo", frame)
# if patch_video_writer is None:
# patch_video_writer = cv2.VideoWriter('patch.avi', cv2.VideoWriter.fourcc('M', 'J', 'P', 'G'), 50,
# (224, 224))
# feature_video_writer = cv2.VideoWriter('features.avi', cv2.VideoWriter.fourcc('M', 'J', 'P', 'G'), 50,
# (224, 224))
# offset = int(max(gt_bbox[2], gt_bbox[3]) / 2)
# x_mid = int(gt_bbox[0] + gt_bbox[2] / 2)
# y_mid = int(gt_bbox[1] + gt_bbox[3] / 2)
# patch = cv2.resize(frame[y_mid - offset:y_mid + offset, x_mid - offset: x_mid + offset, :], (224, 224))
# patch_video_writer.write(patch)
#
# features = cv2.resize(
cap = cv2.VideoCapture(0)
while True:
ret, frame = cap.read()
if ret:
image = preprocess(frame)
img = tf.constant(image)
pred_bbox = infer(img)
for key, value in pred_bbox.items():
boxes = value[:, :, 0:4]
pred_conf = value[:, :, 4:]
boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
scores=tf.reshape(
pred_conf,
(tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
max_output_size_per_class=50,
max_total_size=50,
iou_threshold=0.45,
score_threshold=0.25)
pred_bbox = [
boxes.numpy(),
scores.numpy(),
classes.numpy(),
valid_detections.numpy()
]
image = draw_bbox(frame, pred_bbox)
cv2.imshow('frame', frame)
if cv2.waitKey(1) == ord('q'):
break
frame_size = frame.shape[:2]
image_data = utils.image_preprocess(np.copy(frame),
[input_size, input_size])
image_data = image_data[np.newaxis, ...]
pred_sbbox, pred_mbbox, pred_lbbox = sess.run(
[return_tensors[1], return_tensors[2], return_tensors[3]],
feed_dict={return_tensors[0]: image_data})
pred_bbox = np.concatenate([
np.reshape(pred_sbbox, (-1, 5 + num_classes)),
np.reshape(pred_mbbox, (-1, 5 + num_classes)),
np.reshape(pred_lbbox, (-1, 5 + num_classes))
],
axis=0)
bboxes = utils.postprocess_boxes(pred_bbox, frame_size, input_size,
0.4)
bboxes = utils.nms(bboxes, 0.3, method='nms')
image = utils.draw_bbox(frame, bboxes)
out.write(image)
result = np.asarray(image)
success, frame = vid.read()
num_frame += 1
print("number of frame: ", num_frame)
vid.release()
out.release()
print("end of program")