mmdetection tools/visdrone datasets

[huitang96]

can you share the coco format visdrone dataset? when I convert the visdrone, encounter many problems.

[we1pingyu]

I dont have visdrone dataset on my disk currently. The only thing I can find is a script to convert vd to coco. Please check the data path and feel free to leave comments.

import argparse
import json
import os
import sys
import numpy as np
from PIL import Image
import csv

img_id = 0
ann_id = 0
category_dict = {}
category_instancesonly = {
    'pedestrian': 1,
    'people': 2,
    'bicycle': 3,
    'car': 4,
    'van': 5,
    'truck': 6,
    'tricycle': 7,
    'awning-tricycle': 8,
    'bus': 9,
    'motor': 10,
}
category_id = {
    0: 'ig',
    1: 'pedestrian',
    2: 'people',
    3: 'bicycle',
    4: 'car',
    5: 'van',
    6: 'truck',
    7: 'tricycle',
    8: 'awning-tricycle',
    9: 'bus',
    10: 'motor',
    11: 'others',
}
ann_dict = {}
images = []
annotations = []

data_root = '/data/VisDrone2019-DET-train'
for root, dirs, files in os.walk(os.path.join(data_root, 'annotations')):
    for file in files:
        if file.endswith('.txt'):
            file_path = os.path.join(root, file)
            if len(images) % 50 == 0:
                print("Processed %s images, %s annotations" % (
                    len(images), len(annotations)))
            with open(file_path, "r") as f:
                img = Image.open(
                    os.path.join(data_root, 'images', file.split('.')[0] + '.jpg'))
                width, height = img.size[0], img.size[1]
                image = {}
                image['id'] = img_id
                img_id += 1
                image['width'] = width
                image['height'] = height
                image['file_name'] = file.split('.')[0] + '.jpg'
                images.append(image)
                for line in f.readlines():
                    line = line.strip('\n')
                    x, y, w, h, score, cls = float(line.split(',')[0]), float(line.split(',')[1]), float(
                        line.split(',')[2]), float(line.split(',')[3]), float(line.split(',')[4]), int(
                        line.split(',')[5])
                    if int(cls) not in category_id.keys():
                        continue
                    cls = category_id[int(cls)]
                    if cls not in category_instancesonly:
                        continue  # skip non-instance categories
                    xywh_box = (x, y, w, h)
                    ann = {}
                    ann['id'] = ann_id
                    ann_id += 1
                    ann['image_id'] = image['id']
                    ann['area'] = w * h
                    ann['iscrowd'] = 0
                    ann['bbox'] = xywh_box
                    ann['category_id'] = category_instancesonly[cls]
                    annotations.append(ann)
ann_dict['images'] = images
categories = [{"id": category_instancesonly[name], "name": name} for name in category_instancesonly]
ann_dict['categories'] = categories
ann_dict['annotations'] = annotations
print("Num categories: %s" % len(categories))
print("Num images: %s" % len(images))
print("Num annotations: %s" % len(annotations))
with open(os.path.join(data_root, 'coco_annotations', 'train.json'), 'w') as outfile:
    outfile.write(json.dumps(ann_dict))