def visualize_shanghaitech_keepfull():
HARD_CODE = HardCodeVariable()
shanghaitech_data = ShanghaiTechDataPath(root=HARD_CODE.SHANGHAITECH_PATH)
shanghaitech_data_part_a_train = shanghaitech_data.get_a().get_train().get(
)
saved_folder = "visualize/test_dataloader_shanghaitech"
os.makedirs(saved_folder, exist_ok=True)
train_list, val_list = get_train_val_list(shanghaitech_data_part_a_train,
test_size=0.2)
test_list = None
train_loader, val_loader, test_loader = get_dataloader(
train_list,
val_list,
test_list,
dataset_name="shanghaitech_keepfull",
visualize_mode=True)
# do with train loader
train_loader_iter = iter(train_loader)
for i in range(10):
img, label = next(train_loader_iter)
save_img(img, os.path.join(saved_folder,
"train_img" + str(i) + ".png"))
save_density_map(
label.numpy()[0][0],
os.path.join(saved_folder, "train_label" + str(i) + ".png"))
def visualize_shanghaitech_nonoverlap_downsample():
HARD_CODE = HardCodeVariable()
shanghaitech_data = ShanghaiTechDataPath(root=HARD_CODE.SHANGHAITECH_PATH)
shanghaitech_data_part_a_train = shanghaitech_data.get_a().get_train().get(
)
saved_folder = "visualize/test_dataloader"
os.makedirs(saved_folder, exist_ok=True)
DATA_PATH = HARD_CODE.SHANGHAITECH_PATH_PART_B
train_list, val_list = get_train_val_list(shanghaitech_data_part_a_train,
test_size=0.2)
test_list = None
# create data loader
train_loader, val_loader, test_loader = get_dataloader(
train_list,
val_list,
test_list,
dataset_name="shanghaitech_non_overlap_downsample",
visualize_mode=True)
img, label = next(iter(train_loader))
print(img.shape)
save_img(img, os.path.join(saved_folder,
"overlap_downsample_loader_1.png"))
save_density_map(
label[0].numpy()[0],
os.path.join(saved_folder,
"overlap_downsample_loader_with_p_density1.png"))
print("count1 ", label.numpy()[0].sum())
print("count2 ", label.numpy()[0].sum())
print("count3 ", label.numpy()[0].sum())
print("s1 ", label.shape)
def visualize_shanghaitech_pacnn_with_perspective():
HARD_CODE = HardCodeVariable()
shanghaitech_data = ShanghaiTechDataPath(root=HARD_CODE.SHANGHAITECH_PATH)
shanghaitech_data_part_a_train = shanghaitech_data.get_a().get_train().get(
)
saved_folder = "visualize/test_dataloader"
os.makedirs(saved_folder, exist_ok=True)
DATA_PATH = HARD_CODE.SHANGHAITECH_PATH_PART_A
train_list, val_list = get_train_val_list(shanghaitech_data_part_a_train,
test_size=0.2)
test_list = None
# create data loader
train_loader, val_loader, test_loader = get_dataloader(
train_list, val_list, test_list, dataset_name="ucf_cc_50")
train_loader_pacnn = torch.utils.data.DataLoader(ListDataset(
train_list,
shuffle=True,
transform=transforms.Compose([transforms.ToTensor()]),
train=True,
batch_size=1,
num_workers=4,
dataset_name="shanghaitech_pacnn_with_perspective",
debug=True),
batch_size=1,
num_workers=4)
img, label = next(iter(train_loader_pacnn))
print(img.shape)
save_img(img, os.path.join(saved_folder, "pacnn_loader_img.png"))
save_density_map(
label[0].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_with_p_density1.png"))
save_density_map(
label[1].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_with_p_density2.png"))
save_density_map(
label[2].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_with_p_density3.png"))
save_density_map(label[3].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_p_s_4.png"))
save_density_map(label[4].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_p_5.png"))
print("count1 ", label[0].numpy()[0].sum())
print("count2 ", label[1].numpy()[0].sum())
print("count3 ", label[2].numpy()[0].sum())
print("count4 ", label[3].numpy()[0].sum())
print("count5 ", label[4].numpy()[0].sum())
print("s1 ", label[0].shape)
print("s2 ", label[1].shape)
print("s3 ", label[2].shape)
print("s4 ", label[3].shape)
print("s5 ", label[4].shape)
def visualize_evaluation_shanghaitech_keepfull(
path=None,
dataset="shanghaitech_keepfull_r50",
output="visualize/verify_dataloader_shanghaitech",
meta_data="data_info.txt"):
HARD_CODE = HardCodeVariable()
if path == None:
shanghaitech_data = ShanghaiTechDataPath(
root=HARD_CODE.SHANGHAITECH_PATH)
shanghaitech_data_part_a_train = shanghaitech_data.get_a().get_train(
).get()
path = shanghaitech_data_part_a_train
saved_folder = output
os.makedirs(saved_folder, exist_ok=True)
train_list, val_list = get_train_val_list(path, test_size=0.2)
test_list = None
train_loader, val_loader, test_loader = get_dataloader(
train_list,
val_list,
test_list,
dataset_name=dataset,
visualize_mode=True,
debug=True)
# do with train loader
train_loader_iter = iter(train_loader)
f = open(meta_data, "w")
total = len(train_loader)
for i in range(len(train_loader)):
img, label, debug_data = next(train_loader_iter)
p_count = debug_data["p_count"]
name = debug_data["name"][0]
item_number = img_name_to_int(name)
density_map_count = label.sum()
log_str = str(item_number) + " " + str(
density_map_count.item()) + " " + str(p_count.item())
print(log_str)
f.write(log_str + "\n")
save_img(
img,
os.path.join(saved_folder,
"train_img_" + str(item_number) + ".png"))
save_path = os.path.join(saved_folder,
"train_label_" + str(item_number) + ".png")
save_density_map(label.numpy()[0][0], save_path)
print(str(i) + "/" + str(total))
f.close()
def visualize_evaluation_shanghaitech_keepfull(model):
model = model.cuda()
model.eval()
HARD_CODE = HardCodeVariable()
shanghaitech_data = ShanghaiTechDataPath(root=HARD_CODE.SHANGHAITECH_PATH)
shanghaitech_data_part_a_train = shanghaitech_data.get_a().get_train().get(
)
saved_folder = "visualize/evaluation_dataloader_shanghaitech"
os.makedirs(saved_folder, exist_ok=True)
train_list, val_list = get_train_val_list(shanghaitech_data_part_a_train,
test_size=0.2)
test_list = None
train_loader, val_loader, test_loader = get_dataloader(
train_list,
val_list,
test_list,
dataset_name="shanghaitech_keepfull",
visualize_mode=False,
debug=True)
# do with train loader
train_loader_iter = iter(train_loader)
for i in range(10):
img, label, count = next(train_loader_iter)
# save_img(img, os.path.join(saved_folder, "train_img_" + str(i) +".png"))
save_path = os.path.join(saved_folder,
"train_label_" + str(i) + ".png")
save_pred_path = os.path.join(saved_folder,
"train_pred_" + str(i) + ".png")
save_density_map(label.numpy()[0][0], save_path)
pred = model(img.cuda())
predicted_density_map = pred.detach().cpu().clone().numpy()
predicted_density_map_enlarge = cv2.resize(
np.squeeze(predicted_density_map[0][0]),
(int(predicted_density_map.shape[3] * 8),
int(predicted_density_map.shape[2] * 8)),
interpolation=cv2.INTER_CUBIC) / 64
save_density_map(predicted_density_map_enlarge, save_pred_path)
print("pred " + save_pred_path + " value " +
str(predicted_density_map.sum()))
print("cont compare " + str(predicted_density_map.sum()) + " " +
str(predicted_density_map_enlarge.sum()))
print("shape compare " + str(predicted_density_map.shape) + " " +
str(predicted_density_map_enlarge.shape))
def visualize_ucf_cc_50_pacnn():
HARD_CODE = HardCodeVariable()
saved_folder = "visualize/test_dataloader"
os.makedirs(saved_folder, exist_ok=True)
DATA_PATH = HARD_CODE.UCF_CC_50_PATH
train_list, val_list = get_train_val_list(DATA_PATH, test_size=0.2)
test_list = None
# create data loader
train_loader, val_loader, test_loader = get_dataloader(
train_list, val_list, test_list, dataset_name="ucf_cc_50")
train_loader_pacnn = torch.utils.data.DataLoader(ListDataset(
train_list,
shuffle=True,
transform=transforms.Compose([transforms.ToTensor()]),
train=True,
batch_size=1,
num_workers=4,
dataset_name="shanghaitech_pacnn",
debug=True),
batch_size=1,
num_workers=4)
img, label = next(iter(train_loader_pacnn))
print(img.shape)
save_img(img, os.path.join(saved_folder, "pacnn_loader_img.png"))
save_density_map(label[0].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_density1.png"))
save_density_map(label[1].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_density2.png"))
save_density_map(label[2].numpy()[0],
os.path.join(saved_folder, "pacnn_loader_density3.png"))
print("count1 ", label[0].numpy()[0].sum())
print("count2 ", label[1].numpy()[0].sum())
print("count3 ", label[2].numpy()[0].sum())
from ignite.metrics import Loss, MeanAbsoluteError, MeanSquaredError
from crowd_counting_error_metrics import CrowdCountingMeanAbsoluteError, CrowdCountingMeanSquaredError
import torch
from torch import nn
from models import CSRNet
import os
if __name__ == "__main__":
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
args = real_args_parse()
print(args)
DATA_PATH = args.input
# create list
train_list, val_list = get_train_val_list(DATA_PATH, test_size=0.2)
test_list = None
# create data loader
train_loader, val_loader, test_loader = get_dataloader(
train_list, val_list, test_list, dataset_name="ucf_cc_50")
# model
model = CSRNet()
model = model.to(device)
# loss function
loss_fn = nn.MSELoss(size_average=False).cuda()
optimizer = torch.optim.SGD(model.parameters(),
args.lr,
from torch import nn
from models import CSRNet
import os
if __name__ == "__main__":
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
args = real_args_parse()
print(args)
DATA_PATH = args.input
TRAIN_PATH = os.path.join(DATA_PATH, "train_data")
TEST_PATH = os.path.join(DATA_PATH, "test_data")
# create list
train_list, val_list = get_train_val_list(TRAIN_PATH)
test_list = create_training_image_list(TEST_PATH)
# create data loader
train_loader, val_loader, test_loader = get_dataloader(train_list, val_list, test_list)
# model
model = CSRNet()
model = model.to(device)
# loss function
loss_fn = nn.MSELoss(size_average=False).cuda()
optimizer = torch.optim.SGD(model.parameters(), args.lr,
momentum=args.momentum,
def visualize_evaluation_shanghaitech_keepfull(model, args):
"""
:param model: model with param, if not model then do not output pred
:param args:
:return:
"""
if model is not None:
model = model.cuda()
model.eval()
saved_folder = args.output
os.makedirs(saved_folder, exist_ok=True)
train_list, val_list = get_train_val_list(args.input, test_size=0.2)
test_list = create_image_list(args.input)
train_loader, val_loader, test_loader = get_dataloader(
train_list,
val_list,
test_list,
dataset_name="shanghaitech_keepfull_r50",
visualize_mode=False,
debug=True)
log_f = open(args.meta_data, "w")
mae_s = 0
mse_s = 0
n = 0
train_loader_iter = iter(train_loader)
_, gt_density, _ = next(train_loader_iter)
with torch.no_grad():
for item in test_loader:
img, gt_density, debug_info = item
gt_count = debug_info["p_count"]
file_name = debug_info["name"]
print(file_name[0].split(".")[0])
file_name_only = file_name[0].split(".")[0]
save_path = os.path.join(saved_folder,
"label_" + file_name_only + ".png")
save_pred_path = os.path.join(saved_folder,
"pred_" + file_name_only + ".png")
save_density_map(gt_density.numpy()[0], save_path)
if model is not None:
pred = model(img.cuda())
predicted_density_map = pred.detach().cpu().clone().numpy()
predicted_density_map_enlarge = cv2.resize(
np.squeeze(predicted_density_map[0][0]),
(int(predicted_density_map.shape[3] * 8),
int(predicted_density_map.shape[2] * 8)),
interpolation=cv2.INTER_CUBIC) / 64
save_density_map(predicted_density_map_enlarge, save_pred_path)
print("pred " + save_pred_path + " value " +
str(predicted_density_map.sum()))
print("cont compare " + str(predicted_density_map.sum()) +
" " + str(predicted_density_map_enlarge.sum()))
print("shape compare " + str(predicted_density_map.shape) +
" " + str(predicted_density_map_enlarge.shape))
pred_count = pred.detach().cpu().sum()
pred_count_num = pred_count.item()
error = abs(pred_count_num - gt_count_num)
else:
error = 0
pred_count = 0
mae_s += error
mse_s += error * error
density_map_count = gt_density.detach().sum()
density_map_count_num = density_map_count.item()
gt_count_num = gt_count.item()
if model is not None:
log_str = str(file_name_only) + " " + str(
density_map_count_num) + " " + str(
gt_count.item()) + " " + str(pred_count.item())
else:
log_str = str(file_name_only) + " " + str(
density_map_count_num) + " " + str(gt_count.item())
print(log_str)
log_f.write(log_str + "\n")
log_f.close()
mae = mae_s / n
mse = math.sqrt(mse_s / n)
print("mae ", mae)
print("mse", mse)