experiment.log_text(args.note)
DATA_PATH = args.input
TRAIN_PATH = os.path.join(DATA_PATH, "train_data")
TEST_PATH = os.path.join(DATA_PATH, "test_data")
dataset_name = args.datasetname
if dataset_name=="shanghaitech":
print("will use shanghaitech dataset with crop ")
elif dataset_name == "shanghaitech_keepfull":
print("will use shanghaitech_keepfull")
else:
print("cannot detect dataset_name")
print("current dataset_name is ", dataset_name)
# create list
train_list = create_image_list(TRAIN_PATH)
test_list = create_image_list(TEST_PATH)
# create data loader
train_loader, train_loader_for_eval, test_loader = get_dataloader(train_list, train_list, test_list, dataset_name=dataset_name, batch_size=args.batch_size)
print("len train_loader ", len(train_loader))
# model
model_name = args.model
experiment.log_other("model", model_name)
if model_name == "M1":
model = M1()
elif model_name == "M2":
model = M2()
elif model_name == "M3":
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)
