def take_action(self,action_index):
self.action_trajectory.append(action_index)
result = take_action(self.current_image,action_index)
new_l2_distance = compute_color_l2(current_image=result, target_image=self.target_image)
reward = self.l2_distance_old - new_l2_distance
self.current_image = result
if new_l2_distance < TERMINAL_THRESHOLD:
self.done = True
return result,reward,self.done
def reset(self):
raw_img_path = random.choice(self.raw_path)
raw_img_name = os.path.basename(raw_img_path)
raw_img_index = raw_img_name.split('_')[0]
self.raw_image = load_image(raw_img_path)
self.target_image = load_image( self.target_path+"%s.jpg"%raw_img_index) # 干净的目标图像
self.current_image = self.raw_image
self.l2_distance_old = compute_color_l2(current_image=self.current_image,target_image=self.target_image)
self.action_trajectory = []
self.done = False
return self.current_image
def take_master_action(self, action_index):
self.action_trajectory.append(self.master_action_list[action_index])
result = self.master_action_list[action_index](self.current_image)
new_l2_distance = compute_color_l2(current_image=result, target_image=self.target_image)
reward = self.l2_distance_old - new_l2_distance
self.l2_distance_old = new_l2_distance
self.current_image = result
#print("master_distance: ",new_l2_distance)
if new_l2_distance < TERMINAL_THRESHOLD:
self.done = True
# result = np.clip(result*255.0,a_min=0.0,a_max=255.0).astype(np.uint8)
return result,self.get_color_feature(result),self.get_gray_feature(result),reward,self.done
def take_action(self,action_index,policy_index):
result = self.sub_policy_list[policy_index][action_index](self.current_image)
new_l2_distance = compute_color_l2(current_image=result, target_image=self.target_image)
reward = self.l2_distance_old - new_l2_distance
self.l2_distance_old = new_l2_distance
self.distance = new_l2_distance
self.current_image = result
if new_l2_distance < TERMINAL_THRESHOLD:
self.done = True
# result = np.clip(result*255.0,a_min=0.0,a_max=255.0).astype(np.uint8)
return result,self.get_color_feature(result),self.get_gray_feature(result),reward,self.done
def reset(self):
target_name = random.choice(self.target_name_list)
self.img_name = target_name
index_str = target_name[0:target_name.find('.')]
for name in self.dirty_name_list:
index = name[0:name.find('_')]
if index == index_str:
dirty_name = index
break
dirty_name = dirty_name+"_a.jpg"
# print("dirty_name:%s target_name:%s"%(dirty_name,target_name ))
# print("dirty_path:%s target_path:%s"%(self.dirty_path,self.target_path ))
self.dirty_image = load_image2numpy(self.dirty_path+dirty_name) # 脏数据
self.target_image = load_image2numpy(self.target_path+target_name) # 干净数据
self.current_image = self.dirty_image
self.l2_distance_old = compute_color_l2(current_image=self.dirty_image,target_image=self.target_image)
self.distance = None
self.action_trajectory = []
self.done = False
# result = np.clip(self.current_image*255.0,a_min=0.0,a_max=255.0).astype(np.uint8)
return self.current_image ,self.get_color_feature(self.current_image),self.get_gray_feature(self.current_image)