PASCAL VOC(Visual Object Classes)http://host.robots.ox.ac.uk/pascal/VOC/竞赛项目提供了用于目标分类识别的图片数据集以及development kit(用于访问数据和标签的MATLAB代码),分四种竞赛:
对象分类/识别竞赛(Classification/Detection Competitions)
目标轮廓划分竞赛(Segmentation Competition)
人的动作预测竞赛(Action Classification Competition)
人体各部分识别竞赛(Person Layout Taster Competition)
注:上面四张图片截取自PASCAL VOC网站,我觉得这些图片非常直观的解释了上面的四种任务,懒得用大段的文字说明来解释,所以截图过来在此用作示例。
把VOC2012数据集VOCtrainval_11-May-2012.tar下载下来后展开,可以看到ImageSets目录下对应有四个目录Main、Segmentation、Action、Layout,即分别用于上面四种竞赛:
执行下面命令把PASCAL VOC数据转换成tfrecord格式数据:
cd D:\AI\dataset\VOC2012
python D:\AI\tensorflow\models\research\object_detection\dataset_tools\create_pascal_tf_record.py ^
--label_map_path=D:\AI\tensorflow\models\research\object_detection\data\pascal_label_map.pbtxt ^
--data_dir=D:\AI\dataset ^
--year=VOC2012 ^
--set=train ^
--output_path=pascal_train.record
python D:\AI\tensorflow\models\research\object_detection\dataset_tools\create_pascal_tf_record.py ^
--label_map_path=D:\AI\tensorflow\models\research\object_detection\data\pascal_label_map.pbtxt ^
--data_dir=D:\AI\dataset ^
--year=VOC2012 ^
--set=val ^
--output_path=pascal_val.record
可以看到生成了两个tfrecord数据文件:
把D:\AI\tensorflow\models\research\object_detection\samples\configs\ssd_mobilenet_v1_pets.config拷贝到D:\AI\dataset\VOC2012下并改名为ssd_mobilenet_v1_voc2012.config,修改其内容如下:
# SSD with Mobilenet v1 configuration for MSCOCO Dataset.
# Users should configure the fine_tune_checkpoint field in the train config as
# well as the label_map_path and input_path fields in the train_input_reader and
# eval_input_reader. Search for "PATH_TO_BE_CONFIGURED" to find the fields that
# should be configured.
model {
ssd {
num_classes: 20
box_coder {
faster_rcnn_box_coder {
y_scale: 10.0
x_scale: 10.0
height_scale: 5.0
width_scale: 5.0
}
}
matcher {
argmax_matcher {
matched_threshold: 0.5
unmatched_threshold: 0.5
ignore_thresholds: false
negatives_lower_than_unmatched: true
force_match_for_each_row: true
}
}
similarity_calculator {
iou_similarity {
}
}
anchor_generator {
ssd_anchor_generator {
num_layers: 6
min_scale: 0.2
max_scale: 0.95
aspect_ratios: 1.0
aspect_ratios: 2.0
aspect_ratios: 0.5
aspect_ratios: 3.0
aspect_ratios: 0.3333
}
}
image_resizer {
fixed_shape_resizer {
height: 300
width: 300
}
}
box_predictor {
convolutional_box_predictor {
min_depth: 0
max_depth: 0
num_layers_before_predictor: 0
use_dropout: false
dropout_keep_probability: 0.8
kernel_size: 1
box_code_size: 4
apply_sigmoid_to_scores: false
conv_hyperparams {
activation: RELU_6,
regularizer {
l2_regularizer {
weight: 0.00004
}
}
initializer {
truncated_normal_initializer {
stddev: 0.03
mean: 0.0
}
}
batch_norm {
train: true,
scale: true,
center: true,
decay: 0.9997,
epsilon: 0.001,
}
}
}
}
feature_extractor {
type: 'ssd_mobilenet_v1'
min_depth: 16
depth_multiplier: 1.0
conv_hyperparams {
activation: RELU_6,
regularizer {
l2_regularizer {
weight: 0.00004
}
}
initializer {
truncated_normal_initializer {
stddev: 0.03
mean: 0.0
}
}
batch_norm {
train: true,
scale: true,
center: true,
decay: 0.9997,
epsilon: 0.001,
}
}
}
loss {
classification_loss {
weighted_sigmoid {
}
}
localization_loss {
weighted_smooth_l1 {
}
}
hard_example_miner {
num_hard_examples: 3000
iou_threshold: 0.99
loss_type: CLASSIFICATION
max_negatives_per_positive: 3
min_negatives_per_image: 0
}
classification_weight: 1.0
localization_weight: 1.0
}
normalize_loss_by_num_matches: true
post_processing {
batch_non_max_suppression {
score_threshold: 1e-8
iou_threshold: 0.6
max_detections_per_class: 100
max_total_detections: 100
}
score_converter: SIGMOID
}
}
}
train_config: {
batch_size: 24
optimizer {
rms_prop_optimizer: {
learning_rate: {
exponential_decay_learning_rate {
initial_learning_rate: 0.004
decay_steps: 800720
decay_factor: 0.95
}
}
momentum_optimizer_value: 0.9
decay: 0.9
epsilon: 1.0
}
}
fine_tune_checkpoint: "D:\\AI\\tensorflow\\models\\research\\object_detection\\ssd_mobilenet_v1_coco_2017_11_17\\model.ckpt"
from_detection_checkpoint: true
# Note: The below line limits the training process to 200K steps, which we
# empirically found to be sufficient enough to train the pets dataset. This
# effectively bypasses the learning rate schedule (the learning rate will
# never decay). Remove the below line to train indefinitely.
num_steps: 50000
data_augmentation_options {
random_horizontal_flip {
}
}
data_augmentation_options {
ssd_random_crop {
}
}
}
train_input_reader: {
tf_record_input_reader {
input_path: "D:\\AI\\dataset\\voc2012\\pascal_train.record"
}
label_map_path: "D:\\AI\\tensorflow\\models\\research\\object_detection\\data\\pascal_label_map.pbtxt"
}
eval_config: {
num_examples: 5717
# Note: The below line limits the evaluation process to 10 evaluations.
# Remove the below line to evaluate indefinitely.
max_evals: 10
}
eval_input_reader: {
tf_record_input_reader {
input_path: "D:\\AI\\dataset\\voc2012\\pascal_val.record"
}
label_map_path: "D:\\AI\\tensorflow\\models\\research\\object_detection\\data\\pascal_label_map.pbtxt"
shuffle: false
num_readers: 1
}
根据实际情况需要修改D:\AI\tensorflow\models\research\object_detection\model_main.py,如果你需要打开日志并且需要限制GPU的内存占用的话:
from tensorflow.core.protobuf import config_pb2
tf.logging.set_verbosity(tf.logging.INFO)
...
def main(unused_argv):
flags.mark_flag_as_required('model_dir')
flags.mark_flag_as_required('pipeline_config_path')
gpu_options= config_pb2.GPUOptions(per_process_gpu_memory_fraction=0.5)
session_config = config_pb2.ConfigProto(log_device_placement=True,gpu_options=gpu_options)
config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir,session_config=session_config)
在D:\AI\dataset\VOC2012下面创建train_result和eval_result目录,然后执行下面的命令开始训练或测试/验证:
cd D:\AI\tensorflow\models\research\object_detection
python model_main.py ^
--logtostderr ^
--model_dir=D:\AI\dataset\voc2012\train_result ^
--pipeline_config_path=D:\AI\dataset\voc2012\ssd_mobilenet_v1_voc2012.config ^
--num_train_steps=50000
python model_main.py ^
--logtostderr ^
--model_dir=D:\AI\dataset\voc2012\eval_result ^
--checkpoint_dir=D:\AI\dataset\voc2012\train_result ^
--pipeline_config_path=D:\AI\dataset\voc2012\ssd_mobilenet_v1_voc2012.config ^
--num_eval_steps=5000
我的AI之路(1)--前言
我的AI之路(2)--安装Fedora 28
我的AI之路(3)--安装Anaconda3 和Caffe
我的AI之路(4)--在Anaconda3 下安装Tensorflow 1.8
我的AI之路(5)--如何选择和正确安装跟Tensorflow版本对应的CUDA和cuDNN版本
我的AI之路(6)--在Anaconda3 下安装PyTorch
我的AI之路(7)--安装OpenCV3_Python 3.4.1 + Contrib以及PyCharm
我的AI之路(8)--体验用OpenCV 3的ANN进行手写数字识别及解决遇到的问题
我的AI之路(9)--使用scikit-learn
我的AI之路(10)--如何在Linux下安装CUDA和CUDNN
我的AI之路(11)--如何解决在Linux下编译OpenCV3时出现的多个错误
我的AI之路(12)--如何配置Caffe使用GPU计算并解决编译中出现的若干错误
我的AI之路(13)--解决编译gcc/g++源码过程中出现的错误
我的AI之路(14)--Caffe example:使用MNIST数据集训练和测试LeNet-5模型
我的AI之路(15)--Linux下编译OpenCV3的最新版OpenCV3.4.1及错误解决
我的AI之路(16)--云服务器上安装和调试基于Tensorflow 1.10.1的训练环境
我的AI之路(17)--Tensorflow和Caffe的API及Guide
我的AI之路(18)--Tensorflow的模型安装之object_detection
我的AI之路(19)--如何在Windows下安装pycocotools PythonAPI
我的AI之路(20)--用Tensorflow object_detection跑raccoon数据集
我的AI之路(21)--用Tensorflow object_detection跑PASCAL VOC 2012数据集
我的AI之路(22)--使用Object_Detection_Tensorflow_API
我的AI之路(23)--在Windows下编译Bazel和使用Bazel编译tensorflow