coco_label

最新推荐文章于 2022-10-25 22:39:11 发布

-wxrui-

最新推荐文章于 2022-10-25 22:39:11 发布

阅读量1.5k

点赞数 1

CC 4.0 BY-SA版权

分类专栏：数据集处理

本文链接：https://blog.youkuaiyun.com/qq_42110481/article/details/81366505

本文介绍如何利用COCO数据集训练和应用目标检测模型，涵盖数据预处理、模型选择和评估指标等内容。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

import json
import os
import numpy as np

np.set_printoptions(suppress=True)
json_dir = '/home/wxrui/DATA/coco/coco/annotations'
out_dir = '/home/wxrui/DATA/coco/test'
type_list = ['train', 'val']
year_list = ['2014', '2017']
label_map = {}


def create_label_map(data_dic):
    global label_map
    label_map = data_dic['categories']
    for index, item in enumerate(label_map):
        item['index'] = index
    label_map_file = open(os.path.join(out_dir, 'label_map.json'), 'w')
    json.dump(label_map, label_map_file, indent=4)


def id_2_index(id_list):
    index_list = []
    map_list = [0 for i in range(100)]
    for item in label_map:
        index = item['index']
        id = item['id']
        map_list[id] = index
    for id in id_list:
        index_list.append(map_list[int(id)])
    return np.array(index_list)


def format_label(label, img_width, img_height):
    centerx = label[:, 1] + label[:, 3] / 2
    centery = label[:, 2] + label[:, 4] / 2
    width