【yolov5】用BDD100K数据集制作yolov5人车检测数据集

一、下载BDD100K数据集

数据集下载地址：http://bdd-data.berkeley.edu

二、json格式转yolo的txt格式

下面是一个bdd100k到yolov5的标注转换代码。其中把’car’,‘bus’,'truck’这三个类合并为了一类，“pedestrian”,“other person”，"rider"作为另一类，并过滤掉了晚上的数据
测试集：

import re 
import os 
import json 
 
def search_file(data_dir, pattern=r'\.jpg$'):
    root_dir = os.path.abspath(data_dir)
    for root, dirs, files in os.walk(root_dir):
        for f in files:
            if re.search(pattern, f, re.I):
                abs_path = os.path.join(root, f)
                # print('new file %s' % absfn)
                yield abs_path
 
class Bdd2yolov5:
    def __init__(self):
        self.bdd100k_width = 1280
        self.bdd100k_height = 720
        self.select_categorys=["pedestrian","other person","rider", "car", "bus", "truck"]
        self.cat2id = {
            "pedestrian": 0,
            "other person": 0,
            "rider": 0,
            "car": 1,
            "bus": 1,
            "truck": 1
        } 
 
    def _filter_by_attr(self, attr=None):
        if attr is None:
            return False 
        #过滤掉晚上的图片
        if attr['timeofday'] == 'night':
            return True 
        return False 
 
    def bdd2yolov5(self, path):
        with open(path) as fp:
            fp = json.load(fp)
            for j in fp:
                lines = ""
                label_name = j["name"].replace("jpg", "txt")
                image_name = j["name"]
                if self._filter_by_attr(j['attributes']):
                    img_path = os.path.join(r"D:\Download\bdd100k\images\100k\train", image_name)
                    if os.path.exists(img_path):
                        os.remove(img_path)
                    continue
                for fr in j["labels"]:
                    dw = 1.0 / self.bdd100k_width
                    dh = 1.0 / self.bdd100k_height
                    # for obj in fr["objects"]:
                    if fr["category"] in self.select_categorys:
                        idx = self.cat2id[fr["category"]]
                        cx = (fr["box2d"]["x1"] + fr["box2d"]["x2"]) / 2.0
                        cy = (fr["box2d"]["y1"] + fr["box2d"]["y2"]) / 2.0
                        w = fr["box2d"]["x2"] - fr["box2d"]["x1"]
                        h = fr["box2d"]["y2"] - fr["box2d"]["y1"]
                        if w<=0 or h<=0:
                            continue
                        #根据图片尺寸进行归一化
                        cx,cy,w,h = cx*dw,cy*dh,w*dw,h*dh
                        line = f"{idx} {cx:.6f} {cy:.6f} {w:.6f} {h:.6f}\n"
                        lines += line
                if len(lines) != 0:
                    yolo_txt = os.path.join("./train/train", label_name)
                    # yolo_txt = path.replace(".json",".txt")
                    with open(yolo_txt, 'w') as fp2:
                         fp2.writelines(lines)
                if len(lines)==0:
                    img_path = os.path.join(r"D:\Download\bdd100k\images\100k\train", image_name)
                    if os.path.exists(img_path):
                        os.remove(img_path)

 
 
if __name__ == "__main__":
    bdd_label_dir = "./train"
    cvt=Bdd2yolov5()
    for path in search_file(bdd_label_dir, r"\.json$"):
        cvt.bdd2yolov5(path)

如果图片文件数量和标签文件数量不匹配，删掉多出的一部分图片

import os 
def search_file(data_dir, pattern=r'\.jpg$'):
    root_dir = os.path.abspath(data_dir)
    for root, dirs, files in os.walk(root_dir):
        for f in files:
                yield f

bdd_label_dir = r"D:\Download\bdd100k\images\100k\train"
for path in search_file(bdd_label_dir, r"\.jpg$"):
  txtfile = path.replace("jpg", "txt")
  txt_path = os.path.join(r"D:\Download\bdd100k\labels\det_20\train\train", txtfile)
  jpg_path = os.path.join(bdd_label_dir, path)
  if not os.path.exists(txt_path):
    os.remove(jpg_path)

验证集同理，改下文件路径即可

处理完之后：