def images(download_path):
download_file_by_url(cmr_url, download_path, "SA_64x64.zip", "zip")
img_path = os.path.join(download_path, "SA_64x64", "DICOM")
cmr_images = read_dicom_images(img_path,
sort_instance=True,
sort_patient=True)
return cmr_images[:5, ...]
def test_download_file_by_url(param):
url, output_file_name, file_format = param.split(";")
# run twice to test the code when the file exist
download_file_by_url(url, output_directory, output_file_name, file_format)
download_file_by_url(url, output_directory, output_file_name, file_format)
assert os.path.exists(output_directory.joinpath(output_file_name)) is True
def images(download_path):
download_file_by_url(cmr_url, download_path, "SA_64x64.zip", "zip")
img_path = os.path.join(download_path, "SA_64x64", "DICOM")
cmr_dcm_list = read_dicom_dir(img_path,
sort_instance=True,
sort_patient=True)
cmr_images = dicom2arraylist(dicom_patient_list=cmr_dcm_list,
return_patient_id=False)
return cmr_images[:5]
def main():
args = arg_parse()
# ---- setup device ----
device = "cuda" if torch.cuda.is_available() else "cpu"
print("==> Using device " + device)
# ---- setup configs ----
cfg = get_cfg_defaults()
cfg.merge_from_file(args.cfg)
cfg.freeze()
seed.set_seed(cfg.SOLVER.SEED)
# ---- setup logger and output ----
output_dir = os.path.join(cfg.OUTPUT_DIR, cfg.DATASET.NAME, args.output)
os.makedirs(output_dir, exist_ok=True)
logger = lu.construct_logger("gripnet", output_dir)
logger.info("Using " + device)
logger.info(cfg.dump())
# ---- setup dataset ----
download_file_by_url(cfg.DATASET.URL, cfg.DATASET.ROOT, "pose.pt", "pt")
data = torch.load(os.path.join(cfg.DATASET.ROOT, "pose.pt"))
device = torch.device(device)
data = data.to(device)
# ---- setup model ----
print("==> Building model..")
model = GripNet(
cfg.GRIPN.GG_LAYERS, cfg.GRIPN.GD_LAYERS, cfg.GRIPN.DD_LAYERS, data.n_d_node, data.n_g_node, data.n_dd_edge_type
).to(device)
# TODO Visualize model
# ---- setup trainers ----
optimizer = torch.optim.Adam(model.parameters(), lr=cfg.SOLVER.BASE_LR)
# TODO
trainer = Trainer(cfg, device, data, model, optimizer, logger, output_dir)
if args.resume:
# Load checkpoint
print("==> Resuming from checkpoint..")
cp = torch.load(args.resume)
trainer.model.load_state_dict(cp["net"])
trainer.optim.load_state_dict(cp["optim"])
trainer.epochs = cp["epoch"]
trainer.train_auprc = cp["train_auprc"]
trainer.valid_auprc = cp["valid_auprc"]
trainer.train_auroc = cp["train_auroc"]
trainer.valid_auroc = cp["valid_auroc"]
trainer.train_ap = cp["train_ap"]
trainer.valid_ap = cp["valid_ap"]
trainer.train()
def images(download_path):
download_file_by_url(cmr_url, download_path, "SA_64x64.zip", "zip")
img_path = os.path.join(download_path, "SA_64x64_v2.0", "DICOM")
cmr_dcm_list = read_dicom_dir(img_path,
sort_instance=True,
sort_patient=True,
check_series_uid=True)
dcms = []
for i in range(5):
for j in range(len(cmr_dcm_list[i])):
dcms.append(cmr_dcm_list[i][j])
dcm5_list = check_dicom_series_uid(dcms)
cmr_images = dicom2arraylist(dicom_patient_list=dcm5_list,
return_patient_id=False)
return cmr_images
def download(path):
"""Download dataset.
Office-31 source: https://www.cc.gatech.edu/~judy/domainadapt/#datasets_code
Caltech-256 source: http://www.vision.caltech.edu/Image_Datasets/Caltech256/
Data with this library is adapted from: http://www.stat.ucla.edu/~jxie/iFRAME/code/imageClassification.rar
"""
url = "https://github.com/pykale/data/raw/main/images/office"
if not os.path.exists(path):
os.makedirs(path)
for domain_ in OFFICE_DOMAINS:
filename = "%s.zip" % domain_
data_path = os.path.join(path, filename)
if os.path.exists(data_path):
logging.info(f"Data file {filename} already exists.")
continue
else:
data_url = "%s/%s" % (url, filename)
download_file_by_url(data_url, path, filename, "zip")
logging.info(f"Download {data_url} to {data_path}")
logging.info("[DONE]")
return
def baseline_model(download_path):
download_file_by_url(baseline_url, download_path, "baseline.mat", "mat")
return loadmat(os.path.join(download_path, "baseline.mat"))
def pose_data(download_path):
download_file_by_url(pose_url, download_path, "pose.pt", "pt")
return torch.load(os.path.join(download_path, "pose.pt"))
def main():
args = arg_parse()
# ---- setup configs ----
cfg = get_cfg_defaults()
cfg.merge_from_file(args.cfg)
cfg.freeze()
print(cfg)
save_images = cfg.OUTPUT.SAVE_IMAGES
print(f"Save Images: {save_images}")
# ---- initialize folder to store images ----
save_images_location = cfg.OUTPUT.ROOT
print(f"Save Images: {save_images_location}")
if not os.path.exists(save_images_location):
os.makedirs(save_images_location)
# ---- setup dataset ----
base_dir = cfg.DATASET.BASE_DIR
file_format = cfg.DATASET.FILE_FORAMT
download_file_by_url(cfg.DATASET.SOURCE, cfg.DATASET.ROOT,
"%s.%s" % (base_dir, file_format), file_format)
img_path = os.path.join(cfg.DATASET.ROOT, base_dir, cfg.DATASET.IMG_DIR)
images = read_dicom_images(img_path, sort_instance=True, sort_patient=True)
mask_path = os.path.join(cfg.DATASET.ROOT, base_dir, cfg.DATASET.MASK_DIR)
mask = read_dicom_images(mask_path, sort_instance=True)
landmark_path = os.path.join(cfg.DATASET.ROOT, base_dir,
cfg.DATASET.LANDMARK_FILE)
landmark_df = pd.read_csv(
landmark_path, index_col="Subject") # read .csv file as dataframe
landmarks = landmark_df.iloc[:, :6].values
y = landmark_df["Group"].values
y[np.where(
y != 0
)] = 1 # convert to binary classification problem, i.e. no PH vs PAH
# plot the first phase of images
if save_images:
visualize.plot_multi_images(
images[:, 0, ...],
marker_locs=landmarks,
im_kwargs=dict(cfg.IM_KWARGS),
marker_kwargs=dict(cfg.MARKER_KWARGS),
).savefig(str(save_images_location) + "/0)first_phase.png")
# ---- data pre-processing ----
# ----- image registration -----
img_reg, max_dist = reg_img_stack(images.copy(), landmarks)
if save_images:
visualize.plot_multi_images(img_reg[:, 0, ...],
im_kwargs=dict(cfg.IM_KWARGS)).savefig(
str(save_images_location) +
"/1)image_registration")
# ----- masking -----
img_masked = mask_img_stack(img_reg.copy(), mask[0, 0, ...])
if save_images:
visualize.plot_multi_images(img_masked[:, 0, ...],
im_kwargs=dict(cfg.IM_KWARGS)).savefig(
str(save_images_location) +
"/2)masking")
# ----- resize -----
img_rescaled = rescale_img_stack(img_masked.copy(),
scale=1 / cfg.PROC.SCALE)
if save_images:
visualize.plot_multi_images(img_rescaled[:, 0, ...],
im_kwargs=dict(cfg.IM_KWARGS)).savefig(
str(save_images_location) +
"/3)resize")
# ----- normalization -----
img_norm = normalize_img_stack(img_rescaled.copy())
if save_images:
visualize.plot_multi_images(img_norm[:, 0, ...],
im_kwargs=dict(cfg.IM_KWARGS)).savefig(
str(save_images_location) +
"/4)normalize")
# ---- evaluating machine learning pipeline ----
x = img_norm.copy()
trainer = MPCATrainer(classifier=cfg.PIPELINE.CLASSIFIER, n_features=200)
cv_results = cross_validate(trainer,
x,
y,
cv=10,
scoring=["accuracy", "roc_auc"],
n_jobs=1)
print("Averaged training time: {:.4f} seconds".format(
np.mean(cv_results["fit_time"])))
print("Averaged testing time: {:.4f} seconds".format(
np.mean(cv_results["score_time"])))
print("Averaged Accuracy: {:.4f}".format(
np.mean(cv_results["test_accuracy"])))
print("Averaged AUC: {:.4f}".format(np.mean(cv_results["test_roc_auc"])))
# ---- model weights interpretation ----
trainer.fit(x, y)
weights = trainer.mpca.inverse_transform(
trainer.clf.coef_) - trainer.mpca.mean_
weights = rescale_img_stack(
weights, cfg.PROC.SCALE) # rescale weights to original shape
weights = mask_img_stack(weights, mask[0, 0, ...]) # masking weights
top_weights = model_weights.select_top_weight(
weights, select_ratio=0.02) # select top 2% weights
if save_images:
visualize.plot_weights(
top_weights[0][0],
background_img=images[0][0],
im_kwargs=dict(cfg.IM_KWARGS),
marker_kwargs=dict(cfg.WEIGHT_KWARGS),
).savefig(str(save_images_location) + "/5)weights")
def main():
args = arg_parse()
# ---- setup configs ----
cfg = get_cfg_defaults()
cfg.merge_from_file(args.cfg)
cfg.freeze()
print(cfg)
save_figs = cfg.OUTPUT.SAVE_FIG
fig_format = cfg.SAVE_FIG_KWARGS.format
print(f"Save Figures: {save_figs}")
# ---- initialize folder to store images ----
save_figures_location = cfg.OUTPUT.ROOT
print(f"Save Figures: {save_figures_location}")
if not os.path.exists(save_figures_location):
os.makedirs(save_figures_location)
# ---- setup dataset ----
base_dir = cfg.DATASET.BASE_DIR
file_format = cfg.DATASET.FILE_FORAMT
download_file_by_url(cfg.DATASET.SOURCE, cfg.DATASET.ROOT, "%s.%s" % (base_dir, file_format), file_format)
img_path = os.path.join(cfg.DATASET.ROOT, base_dir, cfg.DATASET.IMG_DIR)
patient_dcm_list = read_dicom_dir(img_path, sort_instance=True, sort_patient=True)
images, patient_ids = dicom2arraylist(patient_dcm_list, return_patient_id=True)
patient_ids = np.array(patient_ids, dtype=int)
n_samples = len(images)
mask_path = os.path.join(cfg.DATASET.ROOT, base_dir, cfg.DATASET.MASK_DIR)
mask_dcm = read_dicom_dir(mask_path, sort_instance=True)
mask = dicom2arraylist(mask_dcm, return_patient_id=False)[0][0, ...]
landmark_path = os.path.join(cfg.DATASET.ROOT, base_dir, cfg.DATASET.LANDMARK_FILE)
landmark_df = pd.read_csv(landmark_path, index_col="Subject").loc[patient_ids] # read .csv file as dataframe
landmarks = landmark_df.iloc[:, :-1].values
y = landmark_df["Group"].values
y[np.where(y != 0)] = 1 # convert to binary classification problem, i.e. no PH vs PAH
# plot the first phase of images with landmarks
marker_names = list(landmark_df.columns[1::2])
markers = []
for marker in marker_names:
marker_name = marker.split(" ")
marker_name.pop(-1)
marker_name = " ".join(marker_name)
markers.append(marker_name)
if save_figs:
n_img_per_fig = 45
n_figures = int(n_samples / n_img_per_fig) + 1
for k in range(n_figures):
visualize.plot_multi_images(
[images[i][0, ...] for i in range(k * n_img_per_fig, min((k + 1) * n_img_per_fig, n_samples))],
marker_locs=landmarks[k * n_img_per_fig : min((k + 1) * n_img_per_fig, n_samples), :],
im_kwargs=dict(cfg.PLT_KWS.IM),
marker_cmap="Set1",
marker_kwargs=dict(cfg.PLT_KWS.MARKER),
marker_titles=markers,
image_titles=list(patient_ids[k * n_img_per_fig : min((k + 1) * n_img_per_fig, n_samples)]),
n_cols=5,
).savefig(
str(save_figures_location) + "/0)landmark_visualization_%s_of_%s.%s" % (k + 1, n_figures, fig_format),
**dict(cfg.SAVE_FIG_KWARGS),
)
# ---- data pre-processing ----
# ----- image registration -----
img_reg, max_dist = reg_img_stack(images.copy(), landmarks, landmarks[0])
plt_kawargs = {**{"im_kwargs": dict(cfg.PLT_KWS.IM), "image_titles": list(patient_ids)}, **dict(cfg.PLT_KWS.PLT)}
if save_figs:
visualize.plot_multi_images([img_reg[i][0, ...] for i in range(n_samples)], **plt_kawargs).savefig(
str(save_figures_location) + "/1)image_registration.%s" % fig_format, **dict(cfg.SAVE_FIG_KWARGS)
)
# ----- masking -----
img_masked = mask_img_stack(img_reg.copy(), mask)
if save_figs:
visualize.plot_multi_images([img_masked[i][0, ...] for i in range(n_samples)], **plt_kawargs).savefig(
str(save_figures_location) + "/2)masking.%s" % fig_format, **dict(cfg.SAVE_FIG_KWARGS)
)
# ----- resize -----
img_rescaled = rescale_img_stack(img_masked.copy(), scale=1 / cfg.PROC.SCALE)
if save_figs:
visualize.plot_multi_images([img_rescaled[i][0, ...] for i in range(n_samples)], **plt_kawargs).savefig(
str(save_figures_location) + "/3)resize.%s" % fig_format, **dict(cfg.SAVE_FIG_KWARGS)
)
# ----- normalization -----
img_norm = normalize_img_stack(img_rescaled.copy())
if save_figs:
visualize.plot_multi_images([img_norm[i][0, ...] for i in range(n_samples)], **plt_kawargs).savefig(
str(save_figures_location) + "/4)normalize.%s" % fig_format, **dict(cfg.SAVE_FIG_KWARGS)
)
# ---- evaluating machine learning pipeline ----
x = np.concatenate([img_norm[i].reshape((1,) + img_norm[i].shape) for i in range(n_samples)], axis=0)
trainer = MPCATrainer(classifier=cfg.PIPELINE.CLASSIFIER, n_features=200)
cv_results = cross_validate(trainer, x, y, cv=10, scoring=["accuracy", "roc_auc"], n_jobs=1)
print("Averaged training time: {:.4f} seconds".format(np.mean(cv_results["fit_time"])))
print("Averaged testing time: {:.4f} seconds".format(np.mean(cv_results["score_time"])))
print("Averaged Accuracy: {:.4f}".format(np.mean(cv_results["test_accuracy"])))
print("Averaged AUC: {:.4f}".format(np.mean(cv_results["test_roc_auc"])))
# ---- model weights interpretation ----
trainer.fit(x, y)
weights = trainer.mpca.inverse_transform(trainer.clf.coef_) - trainer.mpca.mean_
weights = rescale_img_stack(weights, cfg.PROC.SCALE) # rescale weights to original shape
weights = mask_img_stack(weights, mask) # masking weights
top_weights = model_weights.select_top_weight(weights, select_ratio=0.02) # select top 2% weights
if save_figs:
visualize.plot_weights(
top_weights[0][0],
background_img=images[0][0],
im_kwargs=dict(cfg.PLT_KWS.IM),
marker_kwargs=dict(cfg.PLT_KWS.WEIGHT),
).savefig(str(save_figures_location) + "/5)weights.%s" % fig_format, **dict(cfg.SAVE_FIG_KWARGS))
def gait(download_path):
download_file_by_url(gait_url, download_path, "gait.mat", "mat")
return loadmat(os.path.join(download_path, "gait.mat"))
def test_get_source_target(source_cfg, target_cfg, valid_ratio, weight_type,
datasize_type, testing_cfg, class_subset):
source_name, source_n_class, source_trainlist, source_testlist = source_cfg.split(
";")
target_name, target_n_class, target_trainlist, target_testlist = target_cfg.split(
";")
n_class = eval(min(source_n_class, target_n_class))
# get cfg parameters
cfg = testing_cfg
cfg.DATASET.SOURCE = source_name
cfg.DATASET.SRC_TRAINLIST = source_trainlist
cfg.DATASET.SRC_TESTLIST = source_testlist
cfg.DATASET.TARGET = target_name
cfg.DATASET.TGT_TRAINLIST = target_trainlist
cfg.DATASET.TGT_TESTLIST = target_testlist
cfg.DATASET.WEIGHT_TYPE = weight_type
cfg.DATASET.SIZE_TYPE = datasize_type
download_file_by_url(
url=url,
output_directory=str(Path(cfg.DATASET.ROOT).parent.absolute()),
output_file_name="video_test_data.zip",
file_format="zip",
)
# test get_source_target
source, target, num_classes = VideoDataset.get_source_target(
VideoDataset(source_name), VideoDataset(target_name), seed, cfg)
assert num_classes == n_class
assert isinstance(source, dict)
assert isinstance(target, dict)
assert isinstance(source["rgb"], VideoDatasetAccess)
assert isinstance(target["rgb"], VideoDatasetAccess)
assert isinstance(source["flow"], VideoDatasetAccess)
assert isinstance(target["flow"], VideoDatasetAccess)
# test get_train & get_test
assert isinstance(source["rgb"].get_train(), torch.utils.data.Dataset)
assert isinstance(source["rgb"].get_test(), torch.utils.data.Dataset)
assert isinstance(source["flow"].get_train(), torch.utils.data.Dataset)
assert isinstance(source["flow"].get_test(), torch.utils.data.Dataset)
# test get_train_valid
train_valid = source["rgb"].get_train_valid(valid_ratio)
assert isinstance(train_valid, list)
assert isinstance(train_valid[0], torch.utils.data.Dataset)
assert isinstance(train_valid[1], torch.utils.data.Dataset)
# test action_multi_domain_datasets
dataset = VideoMultiDomainDatasets(
source,
target,
image_modality=cfg.DATASET.IMAGE_MODALITY,
seed=seed,
config_weight_type=cfg.DATASET.WEIGHT_TYPE,
config_size_type=cfg.DATASET.SIZE_TYPE,
)
assert isinstance(dataset, DomainsDatasetBase)
# test class subsets
if source_cfg == SOURCES[1] and target_cfg == TARGETS[0]:
dataset_subset = VideoMultiDomainDatasets(
source,
target,
image_modality="rgb",
seed=seed,
config_weight_type=cfg.DATASET.WEIGHT_TYPE,
config_size_type=cfg.DATASET.SIZE_TYPE,
class_ids=class_subset,
)
train, valid = source["rgb"].get_train_valid(valid_ratio)
test = source["rgb"].get_test()
dataset_subset._rgb_source_by_split = {}
dataset_subset._rgb_target_by_split = {}
dataset_subset._rgb_source_by_split["train"] = get_class_subset(
train, class_subset)
dataset_subset._rgb_target_by_split[
"train"] = dataset_subset._rgb_source_by_split["train"]
dataset_subset._rgb_source_by_split["valid"] = get_class_subset(
valid, class_subset)
dataset_subset._rgb_source_by_split["test"] = get_class_subset(
test, class_subset)
# Ground truth length of the subset dataset
train_dataset_subset_length = len(
[1 for data in train if data[1] in class_subset])
valid_dataset_subset_length = len(
[1 for data in valid if data[1] in class_subset])
test_dataset_subset_length = len(
[1 for data in test if data[1] in class_subset])
assert len(dataset_subset._rgb_source_by_split["train"]
) == train_dataset_subset_length
assert len(dataset_subset._rgb_source_by_split["valid"]
) == valid_dataset_subset_length
assert len(dataset_subset._rgb_source_by_split["test"]
) == test_dataset_subset_length
assert len(dataset_subset) == train_dataset_subset_length
def test_video_domain_adapter(source_cfg, target_cfg, image_modality,
da_method, testing_cfg, testing_training_cfg):
source_name, source_n_class, source_trainlist, source_testlist = source_cfg.split(
";")
target_name, target_n_class, target_trainlist, target_testlist = target_cfg.split(
";")
# get cfg parameters
cfg = testing_cfg
cfg.DATASET.SOURCE = source_name
cfg.DATASET.SRC_TRAINLIST = source_trainlist
cfg.DATASET.SRC_TESTLIST = source_testlist
cfg.DATASET.TARGET = target_name
cfg.DATASET.TGT_TRAINLIST = target_trainlist
cfg.DATASET.TGT_TESTLIST = target_testlist
cfg.DATASET.IMAGE_MODALITY = image_modality
cfg.DATASET.WEIGHT_TYPE = WEIGHT_TYPE
cfg.DATASET.SIZE_TYPE = DATASIZE_TYPE
cfg.DAN.USERANDOM = False
# download example data
download_file_by_url(
url=url,
output_directory=str(Path(cfg.DATASET.ROOT).parent.absolute()),
output_file_name="video_test_data.zip",
file_format="zip",
)
# build dataset
source, target, num_classes = VideoDataset.get_source_target(
VideoDataset(source_name), VideoDataset(target_name), seed, cfg)
dataset = VideoMultiDomainDatasets(
source,
target,
image_modality=cfg.DATASET.IMAGE_MODALITY,
seed=seed,
config_weight_type=cfg.DATASET.WEIGHT_TYPE,
config_size_type=cfg.DATASET.SIZE_TYPE,
)
# setup feature extractor
if cfg.DATASET.IMAGE_MODALITY in ["rgb", "flow"]:
class_feature_dim = 1024
domain_feature_dim = class_feature_dim
if cfg.DATASET.IMAGE_MODALITY == "rgb":
feature_network = {"rgb": VideoBoringModel(3), "flow": None}
else:
feature_network = {"rgb": None, "flow": VideoBoringModel(2)}
else:
class_feature_dim = 2048
domain_feature_dim = int(class_feature_dim / 2)
feature_network = {
"rgb": VideoBoringModel(3),
"flow": VideoBoringModel(2)
}
# setup classifier
classifier_network = ClassNetVideo(input_size=class_feature_dim,
n_class=num_classes)
train_params = testing_training_cfg["train_params"]
method_params = {}
method = domain_adapter.Method(da_method)
# setup DA method
if method.is_mmd_method():
model = video_domain_adapter.create_mmd_based_video(
method=method,
dataset=dataset,
image_modality=cfg.DATASET.IMAGE_MODALITY,
feature_extractor=feature_network,
task_classifier=classifier_network,
**method_params,
**train_params,
)
else:
critic_input_size = domain_feature_dim
# setup critic network
if method.is_cdan_method():
if cfg.DAN.USERANDOM:
critic_input_size = 1024
else:
critic_input_size = domain_feature_dim * num_classes
critic_network = DomainNetVideo(input_size=critic_input_size)
if da_method == "CDAN":
method_params["use_random"] = cfg.DAN.USERANDOM
model = video_domain_adapter.create_dann_like_video(
method=method,
dataset=dataset,
image_modality=cfg.DATASET.IMAGE_MODALITY,
feature_extractor=feature_network,
task_classifier=classifier_network,
critic=critic_network,
**method_params,
**train_params,
)
ModelTestHelper.test_model(model, train_params)
