f'train/{image_id}.dicom',
path='D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm '
'Workspace/vbd_cxr/7_POC/DICOMs')
# %% --------------------
# extract zip
extract_files(
f"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace"
f"/vbd_cxr/7_POC/DICOMs/{image_id}.dicom.zip",
"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace"
f"/vbd_cxr/7_POC/DICOMs/", True)
# %% --------------------
raw_img_arr = dicom2array(
f"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace/"
f"vbd_cxr/7_POC/DICOMs/{image_id}.dicom")
raw_img_bb_info = get_bb_info(train_df, image_id,
["x_min", "y_min", "x_max", "y_max", "class_id"])
# %% --------------------
# view image as DICOM w/ bounding boxes
bounding_box_plotter(raw_img_arr, image_id, raw_img_bb_info,
get_label_2_color_dict())
# %% --------------------
# transform image using albumentations
augmentor = albumentations.Compose(
[
# augmentation operations
albumentations.augmentations.transforms.ColorJitter(brightness=0.3,
contrast=0.3,
saturation=0.3,
hue=0.3,
gt = pd.read_csv(
"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace/vbd_cxr/2_data_split/512/unmerged/10_percent_holdout/holdout_df.csv"
)
predictions = pd.read_csv(
"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace/vbd_cxr/5_inference_on_holdout_10_percent/0_predictions/holdout_ensemble_classification_object_detection.csv"
)
# %% --------------------
label2color = get_label_2_color_dict()
# %% --------------------
original_image_ids = gt["image_id"].unique()
# %% --------------------
for image_id in original_image_ids[:10]:
img_as_arr = get_image_as_array(
f"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm Workspace/vbd_cxr/input_data/512x512/train/{image_id}.png"
)
# %% --------------------
left = get_bb_info(gt, image_id,
['x_min', 'y_min', 'x_max', 'y_max', "class_id"])
right = get_bb_info(predictions, image_id,
['x_min', 'y_min', 'x_max', 'y_max', "label"])
# %% --------------------
bounding_box_plotter_side_to_side(img_as_arr, image_id, left, right,
"Ground Truth", "Predictions",
label2color)
13: ("Pulmonary fibrosis", "#e75480"),
14: ("No finding", "#ffffff")
}
# %% --------------------
# IMAGE DIR
img_dir = "D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm " \
"Workspace/vbd_cxr/9_data/512/transformed_data/train"
# %% --------------------
# ANNOTATION DIR
train_data = pd.read_csv(
"D:/GWU/4 Spring 2021/6501 Capstone/VBD CXR/PyCharm "
"Workspace/vbd_cxr/1_merger/wbf_merged/90_percent_train/object_detection/95_percent"
"/80_percent/train_df_0.csv")
# %% --------------------
# image_ids = train_data["image_id"].unique()
image_ids = ["e1a4353d3e747a7150cb06cac73f4d6f"]
# shuffle is inplace operation
random.shuffle(image_ids)
for img in image_ids[:10]:
img_array = get_image_as_array(f"{img_dir}/{img}.jpeg")
# get bounding box info
img_bb_info = get_bb_info(train_data, img, ['x_min', 'y_min', 'x_max', 'y_max', "class_id"])
# plot image with bounding boxes
bounding_box_plotter(img_array, img, img_bb_info, label2color, save_title_or_plot="plot")
def rescaler(predicted_df,
df_with_original_dimension,
source_height_col,
source_width_col,
target_height_col,
target_width_col,
columns=[
"x_min", "y_min", "x_max", "y_max", "label",
"confidence_score"
]):
# get the dimensions from data frame containing repeated rows of image id
extracted_dimension_df = extract_dimension_df(df_with_original_dimension)
image_id_arr = []
x_min_arr = []
y_min_arr = []
x_max_arr = []
y_max_arr = []
label_arr = []
score_arr = []
for img in predicted_df["image_id"].unique():
target_width, target_height = extracted_dimension_df.loc[
img, [target_width_col, target_height_col]]
source_width, source_height = extracted_dimension_df.loc[
img, [source_width_col, source_height_col]]
bounding_boxes_info = get_bb_info(predicted_df, img, columns)
# upscale the predicted bounding boxes based on original scale and visualize it
bounding_boxes_info[:, [0, 1, 2, 3]] = resize_bb_w_h(
bounding_boxes_info[:, [0, 1, 2, 3, 4]], source_width,
source_height, target_width, target_height)
for i in range(len(bounding_boxes_info)):
# class 14 is no findings class and should be represented as 0,0,1,1
if bounding_boxes_info[i][4] == 14:
image_id_arr.append(img)
x_min_arr.append(0)
y_min_arr.append(0)
x_max_arr.append(1)
y_max_arr.append(1)
label_arr.append(bounding_boxes_info[i][4])
score_arr.append(bounding_boxes_info[i][5])
else:
image_id_arr.append(img)
x_min_arr.append(bounding_boxes_info[i][0])
y_min_arr.append(bounding_boxes_info[i][1])
x_max_arr.append(bounding_boxes_info[i][2])
y_max_arr.append(bounding_boxes_info[i][3])
label_arr.append(bounding_boxes_info[i][4])
score_arr.append(bounding_boxes_info[i][5])
scaled_data = pd.DataFrame({
"image_id": image_id_arr,
"x_min": x_min_arr,
"y_min": y_min_arr,
"x_max": x_max_arr,
"y_max": y_max_arr,
"label": label_arr,
"confidence_score": score_arr
})
return scaled_data