def face_bb(test_img):
img_height =512
img_width = 512
img_channels = 3
_CONF = 0.60
_IOU = 0.15
coords = 'centroids' # Whether the box coordinates to be used as targets for the model should be in the 'centroids' or 'minmax' format, see documentation
normalize_coords = True
org_height = test_img.shape[0]
org_width = test_img.shape[1]
test_img = cv2.resize(test_img, (512, 512))
test_img_input = np.expand_dims(test_img,axis=0)
y_pred = model.predict(test_img_input)
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=_CONF,
iou_threshold=_IOU,
top_k='all',
input_coords=coords,
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width)
result = y_pred_decoded[0][0]
det_label = result[0]
det_conf = result[1]
det_xmin = result[2]
det_xmax = result[3]
det_ymin = result[4]
det_ymax = result[5]
#Converting to integers as the indexes to images are only integers
bb = [int(det_xmin),int(det_xmax),int(det_ymin),int(det_ymax)]
return bb
def plot(path):
X = cv2.imread(path)
newimg = cv2.resize(X, (300, 300), interpolation=cv2.INTER_CUBIC)
bg_img = np.zeros((300, 480, 3), dtype=np.uint8)
bg_img[:, 90:390] = newimg
for k in range(90):
bg_img[:, k] = newimg[:, 0]
for k in range(90):
bg_img[:, 390 + k] = newimg[:, -1]
bg_img = bg_img.reshape((1, ) + bg_img.shape)
# print bg_img.shape
y_pred = model.predict(bg_img)
# print y_pred
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=0.1,
iou_threshold=0.1,
top_k='all',
input_coords='centroids',
normalize_coords=False,
img_height=None,
img_width=None)
#print y_pred_decoded
plt.figure(figsize=(20, 12))
plt.imshow(bg_img[0])
current_axis = plt.gca()
#classes = ['background', 'car', 'truck', 'pedestrian', 'bicyclist', 'light'] # Just so we can print class names onto the image instead of IDs
classes = [
'background', 'card'
] # Just so we can print class names onto the image instead of IDs
# Draw the predicted boxes in blue
for box in y_pred_decoded[0]:
label = '{}: {:.2f}'.format(classes[int(box[0])], box[1])
current_axis.add_patch(
plt.Rectangle((box[2], box[4]),
box[3] - box[2],
box[5] - box[4],
color='blue',
fill=False,
linewidth=2))
current_axis.text(box[2],
box[4],
label,
size='x-large',
color='white',
bbox={
'facecolor': 'blue',
'alpha': 1.0
})
plt.show()
def predict(model, dg):
for X, filenames in dg:
Y_pred = model.predict(X)
img_height, img_width = X[0].shape[:-1]
Y_pred_decoded = decode_y2(Y_pred,
confidence_thresh=0.5,
iou_threshold=0.4,
top_k='all',
input_coords='centroids',
normalize_coords=True,
img_height=img_height,
img_width=img_width)
for idx, y_pred_decoded in enumerate(Y_pred_decoded):
for bbox in y_pred_decoded:
yield [filenames[idx]] + bbox.tolist()
def predict(model, X):
y_pred = model.predict(X)
print(y_pred)
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=0.2,
iou_threshold=0.4,
top_k='all',
input_coords='centroids',
normalize_coords=False,
img_height=None,
img_width=None)
print(
"Decoded predictions (output format is [class_id, confidence, xmin, xmax, ymin, ymax]):\n"
)
print(y_pred_decoded)
return y_pred_decoded
print("Image:", filenames[i])
print()
print("Ground truth boxes:\n")
print(y_true[i])
# 3: Make a prediction
model
y_pred = model.predict(X)
# 4: Decode the raw prediction `y_pred`
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=0.5,
iou_threshold=0.4,
top_k='all',
input_coords='centroids',
normalize_coords=False,
img_height=None,
img_width=None)
print(
"Decoded predictions (output format is [class_id, confidence, xmin, xmax, ymin, ymax]):\n"
)
print(y_pred_decoded[i])
# 5: Draw the predicted boxes onto the image
plt.figure(figsize=(20, 12))
plt.imshow(X[i])
current_axis = plt.gca()
# TODO: double check model input channel format, BGR or RGB
rgb_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
resized_rgb_frame = cv2.resize(rgb_frame,
(MODEL_INPUT_WIDTH, MODEL_INPUT_HEIGHT))
X = np.array([resized_rgb_frame])
start_time = time.time()
y_pred = model.predict(X)
interval = time.time() - start_time
print("prediction time:", interval)
# Decode the raw prediction `y_pred`
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=0.7,
iou_threshold=0.4,
top_k='all',
input_coords='centroids',
normalize_coords=True,
img_height=INPUT_CAM_HEIGHT,
img_width=INPUT_CAM_WIDTH)
np.set_printoptions(precision=2, suppress=True, linewidth=90)
#print("Decoded predictions (output format is [class_id, confidence, xmin, ymin, xmax, ymax]):\n")
#print(y_pred_decoded)
frame = plot_bboxes(frame, y_pred_decoded[0])
cv2.imshow('Video', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
i = 0 # Which batch item to look at
print("Image:", filenames[i])
print()
print("Ground truth boxes:\n")
print(y_true[i])
# 3: Make a prediction
y_pred = model.predict(X)
# 4: Decode the raw prediction `y_pred`
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=0.7,
iou_threshold=0.2,
top_k='all',
input_coords='centroids',
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width)
np.set_printoptions(precision=2, suppress=True, linewidth=90)
print("Predicted boxes:\n")
print(y_pred_decoded[i])
# 5: Draw the predicted boxes onto the image
plt.figure(figsize=(20, 12))
plt.imshow(X[i])
current_axis = plt.gca()
for box in y_true[i]:
_CONF = 0.01
_IOU = 0.15
for i in range(n_test_samples/batch_size):
X, y, filenames = next(test_generator)
y_pred = model.predict(X)
y_pred_decoded = decode_y2(y_pred,
confidence_thresh=_CONF,
iou_threshold=_IOU,
top_k='all',
input_coords=coords,
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width)
np.set_printoptions(suppress=True)
for i in range(batch_size):
print colored("image %d :" %i, "cyan")
print colored("predicted", "green")
print y_pred_decoded[i]
print colored("ground truth", "red")
print y[i]
print(colored("now predicting...", "yellow"))
_CONF = 0.01
_IOU = 0.15
for i in range(n_test_samples / batch_size):
X, y, filenames = next(test_generator)
y_pred = model.predict(X)
y_pred_decoded = decode_y2(
y_pred,
confidence_thresh=_CONF,
iou_threshold=_IOU,
top_k="all",
input_coords=coords,
normalize_coords=normalize_coords,
img_height=img_height,
img_width=img_width,
)
np.set_printoptions(suppress=True)
for j in range(batch_size):
print(colored("image %d :" % j, "cyan"))
print(colored("predicted", "green"))
print(y_pred_decoded[j])
print(colored("ground truth", "red"))
print(y[j])
save_bb("./output_test/", filenames[i], y_pred_decoded[i])
