paddlepaddle目标检测之水果检测(yolov3_mobilenet_v1)
一、创建项目
(1)进入到https://aistudio.baidu.com/aistudio/projectoverview/public
(2)创建项目
点击添加数据集:找到这两个
然后创建即可。
会生成以下项目:
二、启动环境,选择GPU版本
然后会进入到以下界面
选择的两个压缩包在/home/aistudio/data/下,先进行解压:
!unzip /home/aistudio/data/data15067/fruit.zip
!unzip /home/aistudio/data/data15072/PaddleDetec.zip
之后在左边文件夹就可以看到解压后的内容了:
三、查看fruit-detection中的内容:
其实是类似pascal voc目标检测数据集的格式
(1) Annotations
以第一个apple_65.xml为例:
folder:文件夹名称
filename:图片名称
path:文件地址
size:图片的大小
object:图片中的对象名称以及其的左下角和右上角的坐标。
<annotation> <folder>train</folder> <filename>apple_65.jpg</filename> <path>C:\tensorflow1\models\research\object_detection\images\train\apple_65.jpg</path> <source> <database>Unknown</database> </source> <size> <width>800</width> <height>600</height> <depth>3</depth> </size> <segmented>0</segmented> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>70</xmin> <ymin>25</ymin> <xmax>290</xmax> <ymax>226</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>35</xmin> <ymin>217</ymin> <xmax>253</xmax> <ymax>453</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>183</xmin> <ymin>177</ymin> <xmax>382</xmax> <ymax>411</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>605</xmin> <ymin>298</ymin> <xmax>787</xmax> <ymax>513</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>498</xmin> <ymin>370</ymin> <xmax>675</xmax> <ymax>567</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>333</xmin> <ymin>239</ymin> <xmax>574</xmax> <ymax>463</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>191</xmin> <ymin>350</ymin> <xmax>373</xmax> <ymax>543</ymax> </bndbox> </object> <object> <name>apple</name> <pose>Unspecified</pose> <truncated>0</truncated> <difficult>0</difficult> <bndbox> <xmin>443</xmin> <ymin>425</ymin> <xmax>655</xmax> <ymax>598</ymax> </bndbox> </object> </annotation>
(2)ImageSets
里面只有一个文件夹Main,Main里面有:
分别看下是什么:
val.txt:验证集图片的名称
orange_92
banana_79
apple_94
apple_93
banana_81
banana_94
orange_77
mixed_23
orange_78
banana_85
apple_92
apple_79
apple_84
orange_83
apple_85
mixed_21
orange_91
orange_89
banana_80
apple_78
banana_93
mixed_22
orange_94
apple_83
banana_90
apple_77
orange_79
apple_81
orange_86
orange_95
banana_88
orange_85
orange_80
apple_80
apple_82
mixed_25
apple_88
banana_83
banana_77
banana_84
banana_92
banana_86
apple_87
orange_84
banana_78
orange_93
orange_90
banana_89
orange_82
apple_90
apple_95
banana_82
banana_91
mixed_24
banana_87
apple_91
orange_81
apple_89
apple_86
orange_87
train.txt:训练集图片的名称,这里就不贴了,有点长,与验证集类似
label_list.txt:类别名称
apple
banana
orange
也就是说,水果分类检测目前只是识别三类。
(3) JPEGImages:存储的就是实际的图片了
找一下apple_65.jpg看看
就是这个样子的
(4) create_list.py、label_list.txt、train.txt、val.txt
import os import os.path as osp import re import random devkit_dir = './' years = ['2007', '2012'] def get_dir(devkit_dir, type): return osp.join(devkit_dir, type) def walk_dir(devkit_dir): filelist_dir = get_dir(devkit_dir, 'ImageSets/Main') annotation_dir = get_dir(devkit_dir, 'Annotations') img_dir = get_dir(devkit_dir, 'JPEGImages') trainval_list = [] test_list = [] added = set() for _, _, files in os.walk(filelist_dir): for fname in files: img_ann_list = [] if re.match('train\.txt', fname): img_ann_list = trainval_list elif re.match('val\.txt', fname): img_ann_list = test_list else: continue fpath = osp.join(filelist_dir, fname) for line in open(fpath): name_prefix = line.strip().split()[0] if name_prefix in added: continue added.add(name_prefix) ann_path = osp.join(annotation_dir, name_prefix + '.xml') img_path = osp.join(img_dir, name_prefix + '.jpg') assert os.path.isfile(ann_path), 'file %s not found.' % ann_path assert os.path.isfile(img_path), 'file %s not found.' % img_path img_ann_list.append((img_path, ann_path)) return trainval_list, test_list def prepare_filelist(devkit_dir, output_dir): trainval_list = [] test_list = [] trainval, test = walk_dir(devkit_dir) trainval_list.extend(trainval) test_list.extend(test) random.shuffle(trainval_list) with open(osp.join(output_dir, 'train.txt'), 'w') as ftrainval: for item in trainval_list: ftrainval.write(item[0] + ' ' + item[1] + '\n') with open(osp.join(output_dir, 'val.txt'), 'w') as ftest: for item in test_list: ftest.write(item[0] + ' ' + item[1] + '\n') if __name__ == '__main__': prepare_filelist(devkit_dir, '.')
将标注信息转换为列表进行存储。
label_list.txt:还是那三种类别
train.txt:./JPEGImages/mixed_20.jpg ./Annotations/mixed_20.xml等一系列路径
val.txt:./JPEGImages/orange_92.jpg ./Annotations/orange_92.xml等一系列路径
至此fruit-dections中的内容就是这么多了。
四、查看PaddleDetection中的内容
(1) configs
各种网络的配置文件
找到yolov3_mobilenet_v1_fruit.yml看看
architecture: YOLOv3
train_feed: YoloTrainFeed
eval_feed: YoloEvalFeed
test_feed: YoloTestFeed
use_gpu: true
max_iters: 20000
log_smooth_window: 20
save_dir: output
snapshot_iter: 200
metric: VOC
map_type: 11point
pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/yolov3_mobilenet_v1.tar
weights: output/yolov3_mobilenet_v1_fruit/best_model
num_classes: 3
finetune_exclude_pretrained_params: ['yolo_output']
YOLOv3:
backbone: MobileNet
yolo_head: YOLOv3Head
MobileNet:
norm_type: sync_bn
norm_decay: 0.
conv_group_scale: 1
with_extra_blocks: false
YOLOv3Head:
anchor_masks: [[6, 7, 8], [3, 4, 5], [0, 1, 2]]
anchors: [[10, 13], [16, 30], [33, 23],
[30, 61], [62, 45], [59, 119],
[116, 90], [156, 198], [373, 326]]
norm_decay: 0.
ignore_thresh: 0.7
label_smooth: true
nms:
background_label: -1
keep_top_k: 100
nms_threshold: 0.45
nms_top_k: 1000
normalized: false
score_threshold: 0.01
LearningRate:
base_lr: 0.00001
schedulers:
- !PiecewiseDecay
gamma: 0.1
milestones:
- 15000
- 18000
- !LinearWarmup
start_factor: 0.
steps: 100
OptimizerBuilder:
optimizer:
momentum: 0.9
type: Momentum
regularizer:
factor: 0.0005
type: L2
YoloTrainFeed:
batch_size: 1
dataset:
dataset_dir: dataset/fruit
annotation: fruit-detection/train.txt
use_default_label: false
num_workers: 16
bufsize: 128
use_process: true
mixup_epoch: -1
sample_transforms:
- !DecodeImage
to_rgb: true
with_mixup: false
- !NormalizeBox {}
- !ExpandImage
max_ratio: 4.0
mean: [123.675, 116.28, 103.53]
prob: 0.5
- !RandomInterpImage
max_size: 0
target_size: 608
- !RandomFlipImage
is_mask_flip: false
is_normalized: true
prob: 0.5
- !NormalizeImage
is_channel_first: false
is_scale: true
mean:
- 0.485
- 0.456
- 0.406
std:
- 0.229
- 0.224
- 0.225
- !Permute
channel_first: true
to_bgr: false
batch_transforms:
- !RandomShape
sizes: [608]
with_background: false
YoloEvalFeed:
batch_size: 1
image_shape: [3, 608, 608]
dataset:
dataset_dir: dataset/fruit
annotation: fruit-detection/val.txt
use_default_label: false
YoloTestFeed:
batch_size: 1
image_shape: [3, 608, 608]
dataset:
dataset_dir: dataset/fruit
annotation: fruit-detection/label_list.txt
use_default_label: false
注意标红的地方即可。
(2)contrib
行人检测和车辆检测?暂时不用管
(3)dataset: 各文件夹下有py文件,用于下载数据集的
(4)demo:用于检测结果的示例图片。
(5)docs:
(6)inference: 用于推断的‘?
(7) ppdet:paddlepaddle检测相关文件
(8) requirements.txt:所需的一些依赖
tqdm
docstring_parser @ http://github.com/willthefrog/docstring_parser/tarball/master
typeguard ; python_version >= '3.4'
tb-paddle
tb-nightly
(9)slim:应该是用于压缩模型的
(10) tools:工具
五、进行训练
训练的代码在tools中的train.py
进入到PaddleDection目录下
在终端输入:python -u tools/train.py -c configs/yolov3_mobilenet_v1_fruit.yml --use_tb=True --eval
如果发现错误No module named ppdet,在train.py中加入
import sys
sys.path.append("/home/aistudio/PaddleDetection")即可
最后卡在了这,不过应该是训练完了,在PaddleDection目录下可以看到output文件夹:
里面有一个迭代时产生的权重信息:
六、进行测试一张图片
python -u tools/infer.py -c configs/yolov3_mobilenet_v1_fruit.yml -o weights=/home/aistudio/PaddleDetection/output/yolov3_mobilenet_v1_fruit/model_final --infer_img=demo/orange_71.jpg
会报错没有相关包,输入以下命令安装:
pip install docstring_parser
pip install pycocotools
之后:
去output下看看orange_71.jpg:
检测出来的是orange,准确率:94%。
知道了检测训练的整个流程,那么去手动标注poscal voc格式的数据,那么就可以实现检测自己想要的东西了。 然后也可以去看下相关目标检测的论文,明白其中的原理,看看源码之类的。