如何从带有图像注释的 CSV 文件中提取值并将它们附加到 Python 中 RetinaNet 的新 CSV 文件中？

Question

I'm trying to convert a csv file from the VGG Image Annotator software into a csv file that can be used in RetinaNet. The format I need for RetinaNet training data is: path/to/image.jpg,x1,y1,x2,y2,class_name . This is an example of my CSV file from VIA: +=============+===========+==============+===========+========================================================+===+ | filename | file_size | region_count | region_id | region_shape_attributes | | +=============+===========+==============+===========+========================================================+===+ | img--30.png | 2331731 | 10 | 0 | {"name":"rect","x":65,"y":778,"width":108,"height":65} | | +-------------+-----------+--------------+-----------+--------------------------------------------------------+---+

Basically, I need to pull the x, y, width, and height attributes from within the brackets and append them to a list. This is my python code:

import csv

via_path = 'data/tiled/via.csv'

image_annotations = []

with open(via_path, "r") as f:
    reader = csv.reader(f, delimiter=",")
    for line in reader: 
        if '#' in line[0][0]:
            # bypassing comments in csv
            continue
        filename = line[1][2:-2]
        # strip brackets, split and get only the values we care about, then convert all the string to int 
        top_left_x, top_left_y, width, height = list(map(int,list(map(float, line[4].strip('][').split(',')[1:]))))

        if width == 0 or height == 0:
            continue

        # move from top left and width/height to x and y values
        if top_left_x < 0:
            top_left_x = 1
        if top_left_y < 0:
            top_left_y = 1
        x1 = top_left_x
        x2 = top_left_x + width
        y1 = top_left_y
        y2 = top_left_y + height 

        # TODO didn't add names this time since it is all one class
        name = "bird"

        # create the csv row
        new_row = []
        new_row.append(filename)
        new_row.append(x1)
        new_row.append(y1)
        new_row.append(x2)
        new_row.append(y2)
        new_row.append(name)

        image_annotations.append(new_row)

This code outputs:

ValueError
---> top_left_x, top_left_y, width, height = list(map(int,list(map(float, line[4].strip('][').split(',')[1:]))))
ValueError: not enough values to unpack (expected 4, got 0)

Answer 1

The region_shape_attributes column is a JSON string. You need to parse it in order to get the values it contains.

Python has built-in JSON support:

import json

# ... open CSV file, for each record ...

    shape = json.parse(line[4])

    top_left_x = shape['x']
    top_left_y = shape['y']
    # etc