def __init__(self,
datasets_path_list,
image_size_dims=[720, 576],
neighborhood_size=32,
neighborhood_radius=32,
grid_radius=4,
grid_angle=45,
train_steps=5,
pred_steps=5):
super().__init__()
self.train_steps = train_steps
self.pred_steps = pred_steps
person_data = []
group_data = []
scene_data = []
ground_truth = []
for i, dataset_path in enumerate(datasets_path_list):
obs = np.load(dataset_path + "/obs.npy")
print(dataset_path, ": {}".format(len(obs)))
ground_truth.append(
(model_expected_ouput(np.load(dataset_path + "/pred.npy"),
self.pred_steps)))
raw_data, _ = preprocess(dataset_path + "/pixel_pos.csv")
person_data.append(person_model_input(obs, self.train_steps))
group_data.append(
log_group_model_input(obs, self.train_steps, neighborhood_size,
image_size_dims, neighborhood_radius,
grid_radius, grid_angle,
[1, 1, 1, 1, 1, 1, 1, 1], raw_data))
scene_data += self.get_sceneData(obs, dataset_path)
self.person_data = np.concatenate(person_data,
axis=0) #(num_obs, 8, 2)
self.group_data = np.concatenate(group_data, axis=0) #(num_obs, 8, -1)
self.scene_data = scene_data #(num_obs, 3, 720, 576)
self.ground_truth = np.concatenate(ground_truth,
axis=0) #(num_obs, 12, 2)
#TODO: call utils api to split data into person, group, scene data
# apply transformations
self.transformations = transforms.Compose([transforms.ToTensor()])
pred_9 = check_9.pred
pred_10 = check_10.pred
#input轨迹数据输入
vehicle_input_1 = vehicle_model_input(obs_veh_1, observed_frame_num) #车辆
vehicle_input_2 = vehicle_model_input(obs_veh_2, observed_frame_num)
vehicle_input_3 = vehicle_model_input(obs_veh_3, observed_frame_num)
vehicle_input_4 = vehicle_model_input(obs_veh_4, observed_frame_num)
vehicle_input_5 = vehicle_model_input(obs_veh_5, observed_frame_num)
vehicle_input_6 = vehicle_model_input(obs_veh_6, observed_frame_num)
vehicle_input_7 = vehicle_model_input(obs_veh_7, observed_frame_num)
vehicle_input_8 = vehicle_model_input(obs_veh_8, observed_frame_num)
vehicle_input_9 = vehicle_model_input(obs_veh_9, observed_frame_num)
vehicle_input_10 = vehicle_model_input(obs_veh_10, observed_frame_num)
person_input_1 = person_model_input(obs_1, observed_frame_num) #行人
person_input_2 = person_model_input(obs_2, observed_frame_num)
person_input_3 = person_model_input(obs_3, observed_frame_num)
person_input_4 = person_model_input(obs_4, observed_frame_num)
person_input_5 = person_model_input(obs_5, observed_frame_num)
person_input_6 = person_model_input(obs_6, observed_frame_num)
person_input_7 = person_model_input(obs_7, observed_frame_num)
person_input_8 = person_model_input(obs_8, observed_frame_num)
person_input_9 = person_model_input(obs_9, observed_frame_num)
person_input_10 = person_model_input(obs_10, observed_frame_num)
#行人影响数据
group_circle_1 = circle_group_model_input(obs_1, observed_frame_num,
neighborhood_size, dimensions_1,
neighborhood_radius, grid_radius,
grid_angle, circle_map_weights,
person_input='person_input_'+str(i)
group_circle='group_circle_'+str(i)
gruop_grid_veh2ped='gruop_grid_veh2ped_'+str(i)
vehicle_expect_output='vehicle_expect_output_'+str(i)
expected_ouput='expected_ouput_'+str(i)
locals()[data_dir] = r'C:\Users\asus\Desktop\lstm项目\ss-lstm_0529\ss-lstm_0529\datadut\0'+str(i)
locals()[veh_data], locals()[numveh] = preprocess_vehicle(locals()[data_dir])
locals()[raw_data], locals()[numPeds]= preprocess(locals()[data_dir])
locals()[check_veh] = vdp.veh_DataProcesser(locals()[data_dir], observed_frame_num, predicting_frame_num)
locals()[check] = dp.DataProcesser(locals()[data_dir], observed_frame_num, predicting_frame_num)
locals()[obs_veh] = locals()[check_veh].obs
locals()[obs] = locals()[check].obs
locals()[pred_veh]= locals()[check_veh].pred
locals()[pred] = locals()[check].pred
locals()[vehicle_input] = vehicle_model_input(locals()[obs_veh], observed_frame_num)
locals()[person_input] = person_model_input(locals()[obs], observed_frame_num)
locals()[group_circle] = circle_group_model_input(locals()[obs], observed_frame_num, neighborhood_size, dimensions_1,
neighborhood_radius, grid_radius, grid_angle, circle_map_weights,
locals()[raw_data])
locals()[gruop_grid_veh2ped] = veh2ped_circle_group_model_input(locals()[obs], observed_frame_num, dimensions_1,
veh_neighborhood_size, grid_radius, grid_angle,
locals()[raw_data],locals()[veh_data]) # 圆形区域，若要矩形区域改成veh2ped_grid_model_input
locals()[vehicle_expect_output] = vehicle_model_expected_ouput(locals()[pred_veh], predicting_frame_num) # 期望输出
locals()[expected_ouput] = model_expected_ouput(locals()[pred], predicting_frame_num)
# 矩形区域只写一个做存档
group_grid_1 = group_model_input(obs_1, observed_frame_num, neighborhood_size, dimensions_1, grid_size, raw_data_1)
group_log_3 = log_group_model_input(obs_3, observed_frame_num, neighborhood_size, dimensions_1, neighborhood_radius,
grid_radius, grid_angle, circle_map_weights, raw_data_3) #logmap同样是存档
# print(data_dir_1)
frame_dir_2 = './data/ETHuniv/frames/'
frame_dir_3 = './data/UCYuniv/frames/'
frame_dir_4 = './data/UCYzara01/frames/'
frame_dir_5 = './data/UCYzara02/frames/'
data_str_1 = 'ETHhotel-'
data_str_2 = 'ETHuniv-'
data_str_3 = 'UCYuniv-'
data_str_4 = 'zara01-'
data_str_5 = 'zara02-'
# data_dir_1
raw_data_1, numPeds_1 = preprocess(data_dir_1)
obs_1 = np.load('./data/obs_1.npy')
pred_1 = np.load('./data/pred_1.npy')
img_1 = np.load('./data/img_1.npy')
person_input_1 = person_model_input(obs_1, observed_frame_num)
expected_ouput_1 = model_expected_ouput(pred_1, predicting_frame_num)
group_log_1 = log_group_model_input(obs_1, observed_frame_num, neighborhood_size, dimensions_1, neighborhood_radius,
grid_radius, grid_angle, circle_map_weights, raw_data_1)
group_grid_1 = group_model_input(obs_1, observed_frame_num, neighborhood_size, dimensions_1, grid_size, raw_data_1)
group_circle_1 = circle_group_model_input(obs_1, observed_frame_num, neighborhood_size, dimensions_1,
neighborhood_radius, grid_radius, grid_angle, circle_map_weights, raw_data_1)
# data_dir_2
raw_data_2, numPeds_2 = preprocess(data_dir_2)
obs_2 = np.load('./data/obs_2.npy')
pred_2 = np.load('./data/pred_2.npy')
img_2 = np.load('./data/img_2_resize.npy')
# img_2 = all_image_tensor(frame_dir_2, data_str_2, obs_2, 576, 720)
person_input_2 = person_model_input(obs_2, observed_frame_num)
expected_ouput_2 = model_expected_ouput(pred_2, predicting_frame_num)
data_str_4 = 'zara01-'
data_str_5 = 'zara02-'
# data_dir_1
raw_data_1, numPeds_1 = preprocess(data_dir_1)
print(raw_data_1)
print(numPeds_1)
check = dp.DataProcesser(data_dir_1, observed_frame_num, predicting_frame_num)
#obs_1 = np.load('./data/obs_1.npy')
#pred_1 = np.load('./data/pred_1.npy')
obs_1 = check.obs
pred_1 = check.pred
#img_1 = np.load('./data/img_1.npy')
img_1 = all_image_tensor(data_dir_1, data_str_1, obs_1, img_width_1,
img_height_1)
person_input_1 = person_model_input(obs_1, observed_frame_num)
expected_ouput_1 = model_expected_ouput(pred_1, predicting_frame_num)
group_log_1 = log_group_model_input(obs_1, observed_frame_num,
neighborhood_size, dimensions_1,
neighborhood_radius, grid_radius,
grid_angle, circle_map_weights, raw_data_1)
group_grid_1 = group_model_input(obs_1, observed_frame_num, neighborhood_size,
dimensions_1, grid_size, raw_data_1)
group_circle_1 = circle_group_model_input(obs_1, observed_frame_num,
neighborhood_size, dimensions_1,
neighborhood_radius, grid_radius,
grid_angle, circle_map_weights,
raw_data_1)
'''
# data_dir_2
raw_data_2, numPeds_2 = preprocess(data_dir_2)