投稿
  1. 首页
  2. java

PaddleDetection开发套件 | VOC格式YOLO3训练

code • 2022-4-26 • java • 阅读 96

PaddleDetection开发套件 | VOC格式YOLO3训练,第1张

一、环境搭建
  • PaddlePaddle 2.2
  • OS 64位 *** 作系统
  • Python 3(3.5.1+/3.6/3.7/3.8/3.9)，64位版本
  • pip/pip3(9.0.1+)，64位版本
  • CUDA >= 10.1
  • cuDNN >= 7.6
# CUDA10.1
python -m pip install paddlepaddle-gpu==2.2.0.post101 -f https://www.paddlepaddle.org.cn/whl/linux/mkl/avx/stable.html

# CPU
python -m pip install paddlepaddle -i https://mirror.baidu.com/pypi/simple

# 在您的Python解释器中确认PaddlePaddle安装成功
>>> import paddle
>>> paddle.utils.run_check()

# 确认PaddlePaddle版本
python -c "import paddle; print(paddle.__version__)"

# 克隆PaddleDetection仓库
cd 
git clone https://github.com/PaddlePaddle/PaddleDetection.git

# 安装其他依赖
cd PaddleDetection
pip install -r requirements.txt

# 编译安装paddledet
python setup.py install

安装后确认测试通过：

python ppdet/modeling/tests/test_architectures.py

测试通过后会提示如下信息：

.......
----------------------------------------------------------------------
Ran 7 tests in 12.816s
OK

有一说一，paddle的安装是真的方便，paddledetection也很简单，比mmlab不知道轻松到哪里去了。

二、准备数据

因为我一直沿用VOC格式数据，所以稍加改动就可以做成paddle需要的样子

如果你并未组织VOC格式的数据集，那么我建议你在标注完成后使用：

YOLO5 训练自定义数据集及避坑指南_訢詡的博客-CSDN博客一、项目地址：https://github.com/ultralytics/yolov5https://github.com/ultralytics/yolov5数据集完全可以沿用VOCDevkit格式数据集二、VOCDevkit格式数据|--VOC2007|---Annotations|---ImageSets|----Layout|----Main...http://www.kaotop.com/file/tupian/20220424/xxxxxxxx1.jpg

aaa/bb/ccc/JPEGImages/xxxxxxxx2.jpg

而paddle要求的为：

aaa/bb/ccc/JPEGImages/xxxxxxxx1.jpg aaa/bb/ccc/Annotation/xxxxxxxx1.xml

aaa/bb/ccc/JPEGImages/xxxxxxxx2.jpg aaa/bb/ccc/Annotation/xxxxxxxx2.xml

按Ctrl+Hda打开替换，点击*按钮打开正则表达式

（1）将(aaa/bb/ccc/JPEGImages/([a-z0-9_-]*))替换为$1.jpg $1.xml

（2）将.xml.jpg替换为.xml

（3）将(aaa/bb/ccc/JPEGImages/([a-z0-9_-]*).jpg aaa/bb/ccc/)/JPEGImages([a-z0-9_-]*.xml)替换为$1/Annotation$2

就可以做出paddle需要的形式了。

>>cd dataset/voc/
>>tree
├── create_list.py
├── download_voc.py
├── generic_det_label_list.txt
├── generic_det_label_list_zh.txt
├── label_list.txt
├── trainval.txt
├── test.txt
├── VOCdevkit/VOC2007
│   ├── annotations
│       ├── 001789.xml
│       |   ...
│   ├── JPEGImages
│       ├── 001789.jpg
│       |   ...
│   ├── ImageSets
│       |   ...
├── VOCdevkit/VOC2012
│   ├── Annotations
│       ├── 2011_003876.xml
│       |   ...
│   ├── JPEGImages
│       ├── 2011_003876.jpg
│       |   ...
│   ├── ImageSets
│       |   ...
|   ...
# label_list.txt 是类别名称列表，文件名必须是 label_list.txt。若使用VOC数据集，config文件中use_default_label为true时不需要这个文件
>>cat label_list.txt
aeroplane
bicycle
...

# trainval.txt 是训练数据集文件列表
>>cat trainval.txt
VOCdevkit/VOC2007/JPEGImages/007276.jpg VOCdevkit/VOC2007/Annotations/007276.xml
VOCdevkit/VOC2012/JPEGImages/2011_002612.jpg VOCdevkit/VOC2012/Annotations/2011_002612.xml
...

# test.txt 是测试数据集文件列表
>>cat test.txt
VOCdevkit/VOC2007/JPEGImages/000001.jpg VOCdevkit/VOC2007/Annotations/000001.xml
...

# label_list.txt voc 类别名称列表
>>cat label_list.txt

aeroplane
bicycle
...
三、配置

（1）上面的label_list.txt里面就得换成自己的分类列表

（2）打开/PaddleDetection/configs/yolov3/yolov3_darknet53_270e_voc.yml

_BASE_: [
  '../datasets/voc.yml',
  '../runtime.yml',
  '_base_/optimizer_270e.yml',
  '_base_/yolov3_darknet53.yml',
  '_base_/yolov3_reader.yml',
]

snapshot_epoch: 5
weights: output/yolov3_darknet53_270e_voc/model_final
# set collate_batch to false because ground-truth info is needed
# on voc dataset and should not collate data in batch when batch size
# is larger than 1.
EvalReader:
  collate_batch: false

它会依赖这5个配置文件

（3）打开/PaddleDetection/configs/yolov3/_base_/optimizer_270e.yml

epoch: 20


LearningRate:
  base_lr: 0.001
  schedulers:
  - !PiecewiseDecay
    gamma: 0.1
    milestones:
    - 216
    - 243
  - !LinearWarmup
    start_factor: 0.
    steps: 4000

OptimizerBuilder:
  optimizer:
    momentum: 0.9
    type: Momentum
  regularizer:
    factor: 0.0005
    type: L2

这里修改训练次数和初始学习率

（4）打开/PaddleDetection/configs/yolov3/_base_/yolov3_reader.yml

worker_num: 2

TrainReader:
  inputs_def:
    num_max_boxes: 50
  sample_transforms:
    - Decode: {}
    - Mixup: {alpha: 1.5, beta: 1.5}
    - RandomDistort: {}
    - RandomExpand: {fill_value: [123.675, 116.28, 103.53]}
    - RandomCrop: {}
    - RandomFlip: {}
  batch_transforms:
    - BatchRandomResize: {target_size: [320, 352, 384, 416, 448, 480, 512, 544, 576, 608], random_size: True, random_interp: True, keep_ratio: False}
    - NormalizeBox: {}
    - PadBox: {num_max_boxes: 50}
    - BboxXYXY2XYWH: {}
    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
    - Permute: {}
    - Gt2YoloTarget: {anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]], anchors: [[114,220],[148,252], [168,293], [193,328], [216,377], [218,261], [248,425], [264,341], [313,485]], downsample_ratios: [32, 16, 8]}
  batch_size: 4
  shuffle: true
  drop_last: true
  mixup_epoch: 250
  use_shared_memory: true

EvalReader:
  inputs_def:
    num_max_boxes: 50
  sample_transforms:
    - Decode: {}
    - Resize: {target_size: [608, 608], keep_ratio: False, interp: 2}
    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
    - Permute: {}
  batch_size: 1

TestReader:
  inputs_def:
    image_shape: [3, 608, 608]
  sample_transforms:
    - Decode: {}
    - Resize: {target_size: [608, 608], keep_ratio: False, interp: 2}
    - NormalizeImage: {mean: [0.485, 0.456, 0.406], std: [0.229, 0.224, 0.225], is_scale: True}
    - Permute: {}
  batch_size: 1

主要在这里面修改锚框和batchsize

（5）打开/PaddleDetection/configs/yolov3/_base_/yolov3_darknet53.yml

architecture: YOLOv3
pretrain_weights: https://paddledet.bj.bcebos.com/models/pretrained/DarkNet53_pretrained.pdparams
norm_type: sync_bn

YOLOv3:
  backbone: DarkNet
  neck: YOLOv3FPN
  yolo_head: YOLOv3Head
  post_process: BBoxPostProcess

DarkNet:
  depth: 53
  return_idx: [2, 3, 4]

# use default config
# YOLOv3FPN:

YOLOv3Head:
  anchors: [[114,220],[148,252], [168,293], [193,328], [216,377], [218,261], [248,425], [264,341], [313,485]]
  anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
  loss: YOLOv3Loss

YOLOv3Loss:
  ignore_thresh: 0.7
  downsample: [32, 16, 8]
  label_smooth: false

BBoxPostProcess:
  decode:
    name: YOLOBox
    conf_thresh: 0.005
    downsample_ratio: 32
    clip_bbox: true
  nms:
    name: MultiClassNMS
    keep_top_k: 100
    score_threshold: 0.01
    nms_threshold: 0.45
    nms_top_k: 1000

同样修改锚框

（6）打开/PaddleDetection/configs/datasets/voc.yml

metric: VOC
map_type: 11point
num_classes: 20


TrainDataset:
  !VOCDataSet
    dataset_dir: dataset/voc
    anno_path: trainval.txt
    label_list: label_list.txt
    data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']

EvalDataset:
  !VOCDataSet
    dataset_dir: dataset/voc
    anno_path: test.txt
    label_list: label_list.txt
    data_fields: ['image', 'gt_bbox', 'gt_class', 'difficult']

TestDataset:
  !ImageFolder
    anno_path: dataset/voc/label_list.txt

这里修改num_classes分类数

四、训练

cd PaddleDetection/
python tools/train.py -c configs/yolov3/yolov3_darknet53_270e_voc.yml

即可开始训练。

欢迎分享，转载请注明来源：内存溢出

原文地址: https://outofmemory.cn/langs/726653.html

赞 (0)
打赏 微信扫一扫 微信扫一扫 支付宝扫一扫 支付宝扫一扫
code code 管理员组
0 0
上一篇 2022-04-26
下一篇 2022-04-26

发表评论

登录后才能评论

评论列表（0条）

保存