classes = np.array(classes)
return boxes, classes
if __name__ == "__main__":
DEBUG = False
NPZ_INDEX = 0
DATA_TO_COLLECT = 1000 # we want to get 1000 data points
seed(123)
environment = launch_env()
policy = PurePursuitPolicy(environment)
MAX_STEPS = 500
SAMPLE_FREQ = 10
while NPZ_INDEX < DATA_TO_COLLECT:
obs = environment.reset()
environment.render(segment=True)
rewards = []
nb_of_steps = 0
while NPZ_INDEX < DATA_TO_COLLECT:
action = policy.predict(np.array(obs))
mask[pos[0],pos[1],:] = segmented_im[pos[0],pos[1],:]
return boxes, labels, mask
def plotWithBoundingBoxes(seg_im,boxes,labels):
for i in range(len(labels)):
cv.rectangle(seg_im, (boxes[i][0],boxes[i][1]), (boxes[i][2],boxes[i][3]), (255,255,255),1)
cv.rectangle(seg_im,(boxes[i][0],boxes[i][1]),(boxes[i][0]+10,boxes[i][1]-12),(255,255,255),cv.FILLED)
cv.putText(seg_im,str(labels[i]),(boxes[i][0],boxes[i][1]),cv.FONT_HERSHEY_COMPLEX,0.5,(0,0,0),1)
return seg_im
seed(123)
environment = launch_env()
policy = PurePursuitPolicy(environment)
MAX_STEPS = 500
while True:
obs = environment.reset() # create first segmented image
environment.render(segment=True) # visualize first segmented image
rewards = []
nb_of_steps = 0
while True:
action = policy.predict(np.array(obs)) # predicts action (v and omega using purepursuit controller)
obs, rew, done, misc = environment.step(action) # Gives non-segmented obs as numpy array
segmented_obs = environment.render_obs(True) # Gives segmented obs as numpy array