最近想训练遥感实例分割,纵观博客发现较少相关 iSAID数据集的切分及数据集转换内容,思来想去应该在繁忙之中抽出时间写个详细的教程。
iSAID数据集下载
iSAID数据集链接
下载上述数据集。
百度网盘中的train和val中包含了实例和语义分割标签。
上述过程只能下载标签,原始图像为DOTA,DOTA图像链接
上述下载完毕后,
train图像:1411张原始图像;1411张实例标签;1411张语义标签。
将所有训练图像放置在一起创建iSAID/train/
val图像:458张原始图像;458张实例标签;458张语义标签。
将所有验证图像放置在一起创建iSAID/val/
切图并分割标签
下载切图代码:切图及标签转换
如果不将图像切分,则会造成显存溢出,原因在于图像具有较多实例,以及大分辨率。
根据readme运行split.py,运行时将’–set’,改为 default=“train,val”
此时执行切图运算(时间较长)。
切割完毕后在iSAID_patches文件夹中
train/84087图像数量
val/19024图像数量
第二步:标签生成:
运行preprocess.py。
注:需要安装lycon库,如果失败,在ubuntu命令行执行:
sudo apt-get install cmake build-essential libjpeg-dev libpng-dev
运行完毕后将生成coco格式的大json文件。
转成YOLO格式并训练
利用coco官方API统计一下目标类别:
# -*- coding: utf-8 -*-
# -----------------------------------------------------
# Time : 2023/2/27 11:28
# Auth : Written by zuofengyuan
# File : 统计coco信息.py
# Copyright (c) Shenyang Pedlin Technolofy Co., Ltd.
# -----------------------------------------------------
"""
Description: TODO
"""
from pycocotools.coco import COCO
# 文件路径
dataDir = r'l/'
dataType = 'train2017' #val2017
annFile = '{}/instances_{}.json'.format(dataDir, dataType)
# initialize COCO api for instance annotations
coco_train = COCO(annFile)
# display COCO categories and supercategories
# 显示所有类别
cats = coco_train.loadCats(coco_train.getCatIds())
cat_nms = [cat['name'] for cat in cats]
print('COCO categories:\n{}'.format('\n'.join(cat_nms)) + '\n')
# 统计单个类别的图片数量与标注数量
for cat_name in cat_nms:
catId = coco_train.getCatIds(catNms=cat_name)
if cat_name == "person":
print(catId)
imgId = coco_train.getImgIds(catIds=catId)
annId = coco_train.getAnnIds(imgIds=imgId, catIds=catId, iscrowd=False)
print("{:<15} {:<6d} {:<10d}\n".format(cat_name, len(imgId), len(annId)))
if cat_name == "motorcycle":
print(catId)
imgId = coco_train.getImgIds(catIds=catId)
annId = coco_train.getAnnIds(imgIds=imgId, catIds=catId, iscrowd=False)
print("{:<15} {:<6d} {:<10d}\n".format(cat_name, len(imgId), len(annId)))
# 统计全部的类别及全部的图片数量和标注数量
print("NUM_categories: " + str(len(coco_train.dataset['categories'])))
print("NUM_images: " + str(len(coco_train.dataset['images'])))
print("NUM_annotations: " + str(len(coco_train.dataset['annotations'])))
loading annotations into memory...
Done (t=19.50s)
creating index...
index created!
COCO categories:
Small_Vehicle
Large_Vehicle
plane
storage_tank
ship
Swimming_pool
Harbor
tennis_court
Ground_Track_Field
Soccer_ball_field
baseball_diamond
Bridge
basketball_court
Roundabout
Helicopter
NUM_categories: 15
NUM_images: 28029
NUM_annotations: 704684
根据官方转换代码JSON-yolomask
将coco格式的大json数据转换成总多yolo格式的关键点数据,
更改yaml数据文件:
train: ../JSON2YOLO-master/new_dir/images/train2017 # train images (relative to 'path') 128 images
val: ../JSON2YOLO-master/new_dir/images/train2017 # val images (relative to 'path') 128 images
test: # test images (optional)
# Classes
names:
0: Small_Vehicle
1: Large_Vehicle
2: plane
3: ship
4: Swimming_pool
5: Harbor
6: tennis_court
7: Swimming_pool
8: Ground_Track_Field
9: Soccer_ball_field
10: baseball_diamond
11: Bridge
12: basketball_court
13: Roundabout
14: Helicopter
然后执行更改配置后执行文章来源:https://www.toymoban.com/news/detail-440303.html
python segment/train.py
查看训练图像
注:coco转yolo mask格式链接:
官方链接文章来源地址https://www.toymoban.com/news/detail-440303.html
到了这里,关于YOLOv5训练大规模的遥感实例分割数据集 iSAID从切图到数据集制作及训练的文章就介绍完了。如果您还想了解更多内容,请在右上角搜索TOY模板网以前的文章或继续浏览下面的相关文章,希望大家以后多多支持TOY模板网!