YOLO自定义对象检测器单次训练3天

Question

我正在尝试实现YOLO对象自定义对象检测器。我分别在colab中下载了1000张图像用于火车数据和100张图像用于验证数据。该代码运行良好，除了collab笔记本需要几乎一天的时间才能获得一个纪元。以下是示例代码。同样，我如何使用Google开放式图像（如kaggle）的自定义人类手部数据集，而无需在collab上下载它。

# -*- coding: utf-8 -*-
"""Hnad_detector.ipynb

Automatically generated by Colaboratory.

Original file is located at
    https://colab.research.google.com/drive/1083eIr_O3Dge1pnMgqliZz9zzMqRtaZC
"""

!git clone https://github.com/EscVM/OIDv4_ToolKit.git

!pip install awscli
!pip install urllib3

!pip3 install -r '/content/OIDv4_ToolKit/requirements.txt'

!python '/content/OIDv4_ToolKit/main.py' downloader --classes '/content/OIDv4_ToolKit/classes.txt' --type_csv train --limit 1000

import os
dir = '/content/OID/Dataset/train/'
for folder in os.listdir(dir):
  print('Number for files in '+folder+'=',len(os.listdir(dir+folder)))

import os
dir = '/content/OID/Dataset/train/'

for folder in os.listdir(dir):
  for file in os.listdir(dir+folder+'/Label')[:5]:
    print(dir+folder+'/Label'+'/'+file)

!git clone https://github.com/thehetpandya/OIDv4_annotation_tool.git

!pip install -r OIDv4_annotation_tool/requirements.txt

from subprocess import check_output
source = '/content/OID/Dataset/train'

for folder in os.listdir(source):
  target = f'{source}/{folder}'
  output = check_output(["python", '/content/OIDv4_annotation_tool/OIDv4_to_VOC.py', "--sourcepath" , f"{source}/{folder}", "--dest_path", f"{target}"])

import shutil
for folder in os.listdir('/content/OID/Dataset/train'):
  print(folder,len(os.listdir('/content/OID/Dataset/train/'+folder)))

for folder in os.listdir(dir):
  shutil.rmtree(f'/{dir}/{folder}/Label')

#import shutil
#shutil.rmtree('/content/OID/Dataset/train/Apple')
#shutil.rmtree('/content/OID/Dataset/train/Light switch')
#shutil.rmtree('/content/OID/Dataset/train/Orange')

import os 
print(len(os.listdir('/content/OID/Dataset/train/Human hand')))

!pip install annotated-images

!pip install annotated-images



import os
images = []
xml = []
for file in os.listdir('/content/OID/Dataset/train/Human hand/'):
  if file.endswith(".jpg"):
    imagepath = '/content/OID/Dataset/train/Human hand/{}'.format(file)

    shutil.move(imagepath,'/content/OID/Dataset/train/Human hand/images')
  if file.endswith(".xml"):
   
    imagepath = '/content/OID/Dataset/train/Human hand/{}'.format(file)
    shutil.move(imagepath,'/content/OID/Dataset/train/Human hand/annotations')

for images in os.listdir('/content/OID/Dataset/train/Human hand/images'):
  print(images[0:17])

shutil.move('/content/OID/Dataset/train/Human hand/images','/content/OID/Dataset/train/Human hand/train')
shutil.move('/content/OID/Dataset/train/Human hand/annotations','/content/OID/Dataset/train/Human hand/train')

!python '/content/OIDv4_ToolKit/main.py' downloader --classes '/content/OIDv4_ToolKit/classes.txt' --type_csv train --limit 300

from subprocess import check_output
source = '/content/OID/Dataset/train'

for folder in os.listdir(source):
  target = f'{source}/{folder}'
  output = check_output(["python", '/content/OIDv4_annotation_tool/OIDv4_to_VOC.py', "--sourcepath" , f"{source}/{folder}", "--dest_path", f"{target}"])

import shutil
for folder in os.listdir('/content/OID/Dataset/train'):
  print(folder,len(os.listdir('/content/OID/Dataset/train/'+folder)))

for folder in os.listdir(dir):
  shutil.rmtree(f'/{dir}/{folder}/Label')

import os
images = []
xml = []
for file in os.listdir('/content/OID/Dataset/train/Human hand/'):
  if file.endswith(".jpg"):
    imagepath = '/content/OID/Dataset/train/Human hand/{}'.format(file)

    shutil.move(imagepath,'/content/OID/Dataset/train/Human hand/validation/images')
  if file.endswith(".xml"):
   
    imagepath = '/content/OID/Dataset/train/Human hand/{}'.format(file)
    shutil.move(imagepath,'/content/OID/Dataset/train/Human hand/validation/annotations')

!wget 'https://github.com/OlafenwaMoses/ImageAI/releases/download/essential-v4/pretrained-yolov3.h5'

#!pip install keras==2.2.4
#!pip install tensorflow==1.13.1
!pip install imageai==2.1.5
#!wget 'https://github.com/OlafenwaMoses/ImageAI/releases/download/1.0/resnet50_coco_best_v2.0.1.h5'

import tensorflow as tf

import imageai as imageai

from imageai.Detection.Custom import DetectionModelTrainer

trainer = DetectionModelTrainer()
trainer.setModelTypeAsYOLOv3()
trainer.setDataDirectory(data_directory="/content/OID/Dataset/train/Human hand/")
trainer.setTrainConfig(object_names_array=["Human hand"], batch_size=8, num_experiments=100, train_from_pretrained_model="/content/pretrained-yolov3.h5")
trainer.trainModel()