Convert attribute columns from VGG Image Annotator CSV to lists, then to pandas DataFrame. Draw the bounding boxes on og images and save the new image.

draw_boxes_and_save.py

# folder where the image is already saved (without bounding boxes)
data_dir = 'media/stable_numbers/number_crops/'

# folder where the new image with bounding boxes should be saved
out_dir = 'media/copy_labeled_number_crops/'

# go through the unique file names (because files can have more than 1 bounding box)
for file in new_df.filename.unique():
    
    # load in the image and .loc[] the rows where it's referenced (each row represents 1 bounding box)
    image = cv.imread(f'{data_dir}{file}')
    box_rows = new_df.loc[new_df.filename == file]

    # go through each row (each bounding box)
    for index, row in box_rows.iterrows():
        
        # set the coordinates of this row's bounding box (+1 in each direction)
        top_left = int(row['x']) - 1, int(row['y']) - 1
        bottom_right = int(top_left[0] + row['width'] + 2), int(top_left[1] + row['height'] + 2)
        
        # put the bounding box on the image in green
        cv.rectangle(image, top_left, bottom_right, color=(0, 255, 0), thickness=1)
        
    # save the new image with bounding boxes to the output directory (same file name)
    new = f'{out_dir}{file}'
    cv.imwrite(new, image)
        

# diaplay a blown up version of the last image (optional)
img = Image.fromarray(image)
img.resize((img.size[0]*5, img.size[1]*5))

make_df.py

a = pd.DataFrame(file_attributes_list)
b = pd.DataFrame(shape_attributes_list)
c = pd.DataFrame(region_attributes_list)

new_df = pd.concat([a, b, c], axis=1, ignore_index=False)

new_df.columns = [c for c in a.columns] + [c for c in b.columns] + [c for c in c.columns]

new_df['filename'] = df.filename
new_df['file_size'] = df.file_size
new_df['region_count'] = df.region_count
new_df['region_id'] = df.region_id
new_df['file_size'] = df.file_size

new_df

make_dict_lists.py

import pandas as pd

df = pd.read_csv('YOURCSVHERE')

file_attributes_list = []

for attribute in df.file_attributes.values:
    attribute = attribute.replace('{', '').replace('}', '').split(',')
    this_row = {}
    for a in attribute:
        a = a.replace('"', '')
        a = a.split(':')
        try:
            try:
                this_row.update({str(a[0]) : int(a[1])})
            except:
                this_row.update({str(a[0]) : str(a[1])})
        except:
            this_row.update({str(a[0]) : ''})
    file_attributes_list.append(this_row)

shape_attributes_list = []

for attribute in df.region_shape_attributes.values:
    attribute = attribute.replace('{', '').replace('}', '').split(',')
    this_row = {}
    for a in attribute:
        a = a.replace('"', '')
        a = a.split(':')
        try:
            try:
                this_row.update({str(a[0]) : int(a[1])})
            except:
                this_row.update({str(a[0]) : str(a[1])})
        except:
            this_row.update({str(a[0]) : ''})
    shape_attributes_list.append(this_row)

region_attributes_list = []

for attribute in df.region_attributes.values:
    attribute = attribute.replace('{', '').replace('}', '').split(',')
    this_row = {}
    for a in attribute:
        a = a.replace('"', '')
        a = a.split(':')
        try:
            try:
                this_row.update({str(a[0]) : int(a[1])})
            except:
                this_row.update({str(a[0]) : str(a[1])})
        except:
            this_row.update({str(a[0]) : ''})
    region_attributes_list.append(this_row)