def generate_thunderhill_batches(df, args):
while True:
batch_x = []
batch_y = []
for idx, row in df.iterrows():
steering_angle = row['steering']
speed = row['speed']
brake = row['brake']
throttle = row['throttle']
img = ReadImg(row['center'])
img, steering_angle = RandomShift(img, steering_angle,
args.adjustement)
img, steering_angle = RandomFlip(img, steering_angle)
img, steering_angle = RandomBrightness(img, steering_angle)
img, steering_angle = RandomRotation(img, steering_angle)
img, steering_angle = RandomBlur(img, steering_angle)
# Preproc is after ....
img = Preproc(img)
batch_x.append(np.reshape(img, (1, HEIGHT, WIDTH, DEPTH)))
batch_y.append([steering_angle, throttle, brake])
if len(batch_x) == args.batch:
batch_x, batch_y = shuffle(batch_x, batch_y)
yield np.vstack(batch_x), np.vstack(batch_y)
batch_x = []
batch_y = []
def generate_batches(df, batch_size, basename='data'):
while True:
batch_x = []
batch_y = []
for idx, row in df.iterrows():
camera = np.random.choice(['left', 'center', 'right'])
img = ReadImg('{}/{}'.format(basename, row[camera].strip()))
if camera == 'left':
steering_angle = min(row['steering'] + .20, 1)
elif camera == 'center':
steering_angle = row['steering']
elif camera == 'right':
steering_angle = max(row['steering'] - .20, -1)
img, steering_angle = RandomShift(img, steering_angle)
img, steering_angle = RandomFlip(img, steering_angle)
img, steering_angle = RandomBrightness(img, steering_angle)
img = Preproc(img)
batch_x.append(np.reshape(img, (1, 66, 200, 3)))
batch_y.append([steering_angle])
if len(batch_x) == batch_size:
batch_x, batch_y = shuffle(batch_x, batch_y)
yield np.vstack(batch_x), np.vstack(batch_y)
batch_x = []
batch_y = []
def generate_thunderhill_batches(gen, batch_size):
batch_x = []
batch_y = []
while True:
for img, steering_angle in gen:
for img1, steering_angle1 in RandomFlip(img, steering_angle):
img1, steering_angle1 = RandomBrightness(img1, steering_angle1)
img1, steering_angle1 = RandomShift(img1, steering_angle1)
img1 = Preproc(img1)
if steering_angle1 > 1:
steering_angle1 = 1
if steering_angle1 < -1:
steering_angle1 = -1
font = cv2.FONT_HERSHEY_SIMPLEX
if show_images:
imgp = Preproc(img.copy())
imgshow = (np.concatenate((imgp, img1), axis=1) + 0.5)
cv2.putText(imgshow, '%.3f' % (steering_angle), (10, 50),
font, 1, (0, 0, 255), 1, cv2.LINE_AA)
cv2.putText(imgshow, '%.3f' % (steering_angle1), (330, 50),
font, 1, (0, 0, 255), 1, cv2.LINE_AA)
cv2.imshow('img', imgshow)
cv2.waitKey(0)
batch_x.append(np.reshape(img1, (1, HEIGHT, WIDTH, DEPTH)))
batch_y.append([steering_angle1])
if len(batch_x) == batch_size:
batch_x, batch_y = shuffle(batch_x, batch_y)
yield np.vstack(batch_x), np.vstack(batch_y)
batch_x = []
batch_y = []
def generate_thunderhill_batches(gen, batch_size):
batch_x = []
batch_x2 = []
batch_y = []
tasks=[]
while True:
for img, steering_angle, throttle, brake, speed in gen:
for img1, steering_angle1 in RandomFlip(img, steering_angle):
img1 = Preproc(img1)
tasks.append((img1, steering_angle1, throttle, brake, speed))
if len(tasks)==BATCH_SIZE:
results = dispatch_tasks(worker, tasks, multiprocessing.cpu_count())
tasks=[]
for i,result in enumerate(results):
res_list = result.get()
batch_x.append(res_list[0][0])
batch_x2.append(res_list[0][1])
batch_y.append(res_list[1])
print(len(batch_x))
if len(batch_x) == batch_size:
batch_x, batch_x2, batch_y = shuffle(batch_x, batch_x2, batch_y)
yield [np.vstack(batch_x), np.vstack(batch_x2)], np.vstack(batch_y)
batch_x = []
batch_x2 = []
batch_y = []
def generate_thunderhill_batches(df, args):
while True:
batch_x = []
batch_y = []
sample_weights = []
for idx, row in df.iterrows():
steering_angle = row['steering']
speed = row['speed']
brake = row['brake']
throttle = row['throttle']
longitude = row['longitude']
latitude = row['latitude']
img = ReadImg(row['center'])
img, steering_angle = RandomShift(img, steering_angle,
args.adjustement)
img, steering_angle = RandomFlip(img, steering_angle)
img, steering_angle = RandomBrightness(img, steering_angle)
img, steering_angle = RandomRotation(img, steering_angle)
img, steering_angle = RandomBlur(img, steering_angle)
# Preproc is after ....
img = Preproc(img)
batch_x.append(np.reshape(img, (1, HEIGHT, WIDTH, DEPTH)))
# Lap Distance
UTMp = LatLontoUTM(RadToDeg(longitude), RadToDeg(latitude))
_, _, LapDistance = NearestWayPointCTEandDistance(UTMp)
LapDistance = 2 * (LapDistance / total_lap_distance) - 1
batch_y.append([steering_angle, LapDistance])
sample_weights.append(row['norm'])
if len(batch_x) == args.batch:
yield (np.vstack(batch_x), np.vstack(batch_y),
np.array(sample_weights))
batch_x = []
batch_y = []
sample_weights = []
def generate_thunderhill_batches(df, batch_size):
while True:
batch_x = []
batch_y = []
for idx, row in df.iterrows():
steering_angle = row['steering']
# img = ReadImg('{}/{}'.format(basename, row['center'].strip()))
img = ReadImg(row['center'])
img, steering_angle = RandomShift(img, steering_angle)
img, steering_angle = RandomFlip(img, steering_angle)
img, steering_angle = RandomBrightness(img, steering_angle)
img = Preproc(img)
batch_x.append(np.reshape(img, (1, 66, 200, 3)))
batch_y.append([steering_angle])
if len(batch_x) == batch_size:
batch_x, batch_y = shuffle(batch_x, batch_y)
yield np.vstack(batch_x), np.vstack(batch_y)
batch_x = []
batch_y = []
def generate_thunderhill_batches(gen, batch_size):
batch_x = []
batch_x2 = []
batch_y = []
while True:
for img, steering_angle, throttle, brake, speed, longitude, latitude in gen:
for img1, steering_angle1 in RandomFlip(img, steering_angle):
img1, steering_angle1 = RandomBrightness(img1, steering_angle1)
img1 = Preproc(img1)
img1, steering_angle1, throttle1 = RandomShift(img1, steering_angle1,throttle)
brake1=brake
if steering_angle1>0.5:
brake1=1
throttle1=0
if steering_angle1<-0.5:
brake1=1
throttle1=0
if np.random.uniform() < 0.5:
speed1 = speed * np.random.uniform()
else:
speed1 = speed
# speed_cl=np.array(speedToClass(speed))
# speed_cl1=np.array(speedToClass(speed1))
if speed1 < 20*0.44704:
throttle1=1
if speed1 > 50*0.44704:
throttle1=0
if speed1 > 70*0.44704:
throttle1=0
brake=1
if brake>0.7:
throttle=0
# if speed_cl[-2]==1 and steering_angle1!=0:
# throttle1=0
# if speed_cl[-1]==1 and steering_angle1!=0:
# brake1=1
# throttle1=0
batch_x.append(np.reshape(img1, (1, HEIGHT, WIDTH, DEPTH)))
spd1 = speed1/40 - 0.5
batch_x2.append(spd1)
font = cv2.FONT_HERSHEY_SIMPLEX
if show_images:
imgp = Preproc(img.copy()[::-1,:,:])
imgshow=(np.concatenate((imgp,img1[::-1,:,:]),axis=0)+0.5)*0.5
imgshow = cv2.resize(imgshow,(0,0),fx=2,fy=2)
# cv2.putText(imgshow,'%s'%(str(list(speed_cl))),(10,30), font, 0.8,(0,0,255),2,cv2.LINE_AA)
# cv2.putText(imgshow,'%s'%(str(list(speed_cl1))),(10,190), font, 0.8,(0,0,255),2,cv2.LINE_AA)
cv2.putText(imgshow,'%.3f %.1f %.1f %.1f'%(steering_angle,throttle, brake,speed),(10,140), font, 0.8,(0,0,255),2,cv2.LINE_AA)
cv2.putText(imgshow,'%.3f %.1f %.1f %.1f'%(steering_angle1,throttle1, brake1,speed1),(10,300), font, 0.8,(0,0,255),2,cv2.LINE_AA)
cv2.putText(imgshow,'%.5f %.5f'%(longitude, latitude),(10,110), font, 0.8,(0,0,255),2,cv2.LINE_AA)
cv2.imshow('img',imgshow)
cv2.waitKey(0)
batch_y.append([steering_angle1,throttle1, brake1])
if len(batch_x) == batch_size:
batch_x, batch_x2, batch_y = shuffle(batch_x, batch_x2, batch_y)
yield [np.vstack(batch_x), np.vstack(batch_x2)], np.vstack(batch_y)
batch_x = []
batch_x2 = []
batch_y = []
def generate_thunderhill_batches(gen, batch_size):
batch_x = []
batch_x2 = []
batch_y = []
while True:
for img, steering_angle, throttle, brake, speed in gen:
for img1, steering_angle1 in RandomFlip(img, steering_angle):
img1, steering_angle1 = RandomBrightness(img1, steering_angle1)
img1, steering_angle1, throttle1 = RandomShift(
img1, steering_angle1, throttle)
img1 = Preproc(img1)
if steering_angle1 > 1:
steering_angle1 = 1
if steering_angle1 < -1:
steering_angle1 = -1
brake1 = brake
if steering_angle1 > 0.5:
brake1 = 1
throttle1 = 0
if steering_angle1 < -0.5:
brake1 = 1
throttle1 = 0
speed_cl = np.array(speedToClass(speed))
if speed_cl[-2] == 1 and steering_angle1 != 0:
throttle1 = 0
if speed_cl[-1] == 1 and steering_angle1 != 0:
brake1 = 1
throttle1 = 0
font = cv2.FONT_HERSHEY_SIMPLEX
if show_images:
imgp = Preproc(img.copy())
imgshow = (np.concatenate(
(imgp, img1), axis=0) + 0.5) * 0.5
cv2.putText(imgshow, '%s' % (str(list(speed_cl))),
(10, 30), font, 0.4, (0, 0, 255), 1,
cv2.LINE_AA)
cv2.putText(
imgshow, '%.3f %.1f %.1f %.1f' %
(steering_angle, throttle, brake, speed), (10, 70),
font, 0.4, (0, 0, 255), 1, cv2.LINE_AA)
cv2.putText(
imgshow, '%.3f %.1f %.1f' %
(steering_angle1, throttle1, brake1), (10, 150), font,
0.4, (0, 0, 255), 1, cv2.LINE_AA)
cv2.imshow('img', imgshow)
cv2.waitKey(0)
batch_x.append(np.reshape(img1, (1, HEIGHT, WIDTH, DEPTH)))
batch_x2.append(speed_cl)
batch_y.append([steering_angle1, throttle1, brake1])
if len(batch_x) == batch_size:
batch_x, batch_x2, batch_y = shuffle(
batch_x, batch_x2, batch_y)
yield [np.vstack(batch_x),
np.vstack(batch_x2)], np.vstack(batch_y)
batch_x = []
batch_x2 = []
batch_y = []
def generate_thunderhill_batches(gen, batch_size):
batch_x = []
batch_x2 = []
batch_x3 = []
batch_y = []
while True:
for img, steering_angle, throttle, brake, speed, longitude, latitude in gen:
for img1, steering_angle1 in RandomFlip(img, steering_angle):
img1, steering_angle1 = RandomBrightness(img1, steering_angle1)
img1 = Preproc(img1)
img1, steering_angle1, throttle1 = RandomShift(
img1, steering_angle1, throttle)
brake1 = brake
if steering_angle1 > 0.5:
brake1 = 1
throttle1 = 0
if steering_angle1 < -0.5:
brake1 = 1
throttle1 = 0
if np.random.uniform() < 0.5:
speed1 = speed * np.random.uniform()
else:
speed1 = speed
speed_cl = np.array(speedToClass(speed))
speed_cl1 = np.array(speedToClass(speed1))
if speed1 < 20 * 0.44704:
throttle1 = 1
if speed1 > 50 * 0.44704:
throttle1 = 0
if speed1 > 70 * 0.44704:
throttle1 = 0
brake = 1
if brake > 0.7:
throttle = 0
# if speed_cl[-2]==1 and steering_angle1!=0:
# throttle1=0
# if speed_cl[-1]==1 and steering_angle1!=0:
# brake1=1
# throttle1=0
batch_x.append(np.reshape(img1, (1, HEIGHT, WIDTH, DEPTH)))
batch_x2.append(speed_cl1)
UTMp = LatLontoUTM(RadToDeg(longitude), RadToDeg(latitude))
CTE, Distance2NextWaypoint, LapDistance = NearestWayPointCTEandDistance(
UTMp)
CTE /= 5.0
LapDistance = 2 * (LapDistance / total_lap_distance) - 1
font = cv2.FONT_HERSHEY_SIMPLEX
if show_images and (CTE > 100 or LapDistance > 2
or LapDistance < -2):
imgp = Preproc(img.copy()[::-1, :, :])
imgshow = (np.concatenate(
(imgp, img1[::-1, :, :]), axis=0) + 0.5) * 0.5
imgshow = cv2.resize(imgshow, (0, 0), fx=2, fy=2)
cv2.putText(imgshow, '%s' % (str(list(speed_cl))),
(10, 30), font, 0.8, (0, 0, 255), 2,
cv2.LINE_AA)
cv2.putText(imgshow, '%s' % (str(list(speed_cl1))),
(10, 190), font, 0.8, (0, 0, 255), 2,
cv2.LINE_AA)
cv2.putText(
imgshow, '%.3f %.1f %.1f %.1f' %
(steering_angle, throttle, brake, speed), (10, 140),
font, 0.8, (0, 0, 255), 2, cv2.LINE_AA)
cv2.putText(
imgshow, '%.3f %.1f %.1f %.1f' %
(steering_angle1, throttle1, brake1, speed1),
(10, 300), font, 0.8, (0, 0, 255), 2, cv2.LINE_AA)
cv2.putText(imgshow, '%.5f %.5f' % (longitude, latitude),
(10, 110), font, 0.8, (0, 0, 255), 2,
cv2.LINE_AA)
cv2.putText(imgshow, '%.3f %.3f' % (CTE, LapDistance),
(10, 250), font, 0.8, (0, 0, 255), 2,
cv2.LINE_AA)
cv2.imshow('img', imgshow)
cv2.waitKey(0)
batch_x3.append((CTE, LapDistance))
batch_y.append([steering_angle1, throttle1, brake1])
if len(batch_x) == batch_size:
batch_x, batch_x2, batch_x3, batch_y = shuffle(
batch_x, batch_x2, batch_x3, batch_y)
yield [
np.vstack(batch_x),
np.vstack(batch_x2),
np.vstack(batch_x3)
], np.vstack(batch_y)
batch_x = []
batch_x2 = []
batch_x3 = []
batch_y = []