Tensorflow在更新1.2版本之后多了很多新功能,其中放出了很多用tf框架写的深度网络结构(看这里),大大降低了吾等调包侠的开发难度,无论是fine-tuning还是该网络结构都方便了不少。这里讲的的是物体检测(object detection)API,这个库的说明文档很详细,可以的话直接看原文即可。
这个物体检测API提供了5种网络结构的预训练的weights,全部是用COCO数据集进行训练,可以在这里下载:分别是SSD+mobilenet, SSD+inception_v2, R-FCN+resnet101, faster RCNN+resnet101, faster RCNN+inception+resnet101。各个模型的精度和计算所需时间如下,具体测评细节可以看这篇文章:
依赖包
Protobuf 2.6
Pillow 1.0
lxml
tf Slim
Jupyter notebook
Matplotlib # 用这个画图会比较慢,内存占用高,可以用cv2来代替
Tensorflow
API安装
$ pip install tensorflow-gpu
$ sudo apt-get install protobuf-compiler python-pil python-lxml
$ sudo pip install jupyter
$ sudo pip install matplotlib
因为使用protobuf来配置模型和训练参数,所以API正常使用必须先编译protobuf库:
$ cd tensorflow/models
$ protoc object_detection/protos/*.proto --python_out=.
然后将models和slim(tf高级框架)加入python环境变量:
export PYTHONPATH=$PYTHONPATH:/your/path/to/tensorflow/models:/your/path/to/tensorflow/models/slim
最后测试安装:
python object_detection/builders/model_builder_test.py
fine-tuning
python object_detection/create_pascal_tf_record.py \
--label_map_path=object_detection/data/pascal_label_map.pbtxt \ # 训练物品的品类和id
--data_dir=VOCdevkit --year=VOC2012 --set=train \
--output_path=pascal_train.record
python object_detection/create_pascal_tf_record.py \
--label_map_path=object_detection/data/pascal_label_map.pbtxt \
--data_dir=VOCdevkit --year=VOC2012 --set=val \
--output_path=pascal_val.record
其中--data_dir
为训练集的目录。结构同Pascal VOC,如下:
+ VOCdevkit # +为文件夹
+ JPEGImages
- 001.jpg # - 为文件
+ Annotations
- 001.xml
- 训练
train和eval输入输出数据储存结构为:
+ input
- label_map.pbtxt file # 可以在object_detection/data/*.pbtxt找到样例
- train TFRecord file
- eval TFRecord file
+ models
+ modelA
- pipeline config file # 可以在object_detection/samples/configs/*.config下找到样例,定义训练参数和输入数据
+ train # 保存训练产生的checkpoint文件
+ eval
准备好上述文件后就可以直接调用train文件进行训练
python object_detection/train.py \
--logtostderr \
--pipeline_config_path=/your/path/to/models/modelA/pipeline config file \
--train_dir=/your/path/to/models/modelA/train
- 评估
在训练开始以后,就可以运行eval来评估模型的效果。不过实际情况是eval模型也需要加载ckpt文件,因此也需要占用不小的显存,而一般训练的时候都会调整batch尽量利用显卡性能,所以想要实时运行train和eval的话需要调整好两者所需的内存。
python object_detection/eval.py \
--logtostderr \
--pipeline_config_path=/your/path/to/models/modelA/pipeline config file \
--checkpoint_dir=/your/path/to/models/modelA/train \
--eval_dir=/your/path/to/models/modelA/eval
- 监控
通过tensorboard命令可以在浏览器很轻松的监控训练进程,在浏览器输入localhost:6006
(默认)即可
tensorboard --logdir=/your/path/to/models/modelA # 需要包含eval和train目录(.ckpt, .index, .meta, checkpoint, graph.pbtxt文件)
freeze model
在训练完成后需要将训练产生的最后一组.meta, .index, .ckpt, checkpoint文件。其中meta保存了graph和metadata,ckpt保存了网络的weights。而在生产环境中进行预测的时候是只需要模型和权重,不需要metadata,所以需要将其提出进行freeze操作,将所需的部分放到一个文件,方便之后的调用,也减少模型加载所需的内存。(在下载的预训练模型解压后可以找到4个文件,其中名为frozen_inference_graph.pb的文件就是freeze后产生的模型文件,比weights文件大,但是比weights和meta文件加起来要小不少。)
本来,tensorflow/python/tools/freeze_graph.py
提供了freeze model的api,但是需要提供输出的final node names(一般是softmax之类的最后一层的激活函数命名),而object detection api提供提供了预训练好的网络,final node name并不好找,所以object_detection
目录下还提供了export_inference_graph.py
。
python export_inference_graph.py \
--input_type image_tensor \
--pipeline_config_path /your/path/to/models/modelA/pipeline config file \
--checkpoint_path /your/path/to/models/modelA/train/model.ckpt-* \
--inference_graph_path /your/path/to/models/modelA/train/frozen_inference_graph.pb # 输出的文件名
模型调用
目录下提供了一个样例。这里只是稍作调整用cv2来显示图像。
import numpy as np
import os, sys
import tensorflow as tf
import cv2
MODEL_ROOT = "/home/arkenstone/tensorflow/workspace/models"
sys.path.append(MODEL_ROOT) # 应用和训练的目录在不同的地方
from object_detection.utils import label_map_util
from object_detection.utils import visualization_utils as vis_util
MODEL_PATH = "/home/arkenstone/tensorflow/workspace/models/objectdetection/models/faster_rcnn_inception_resnet_v2_atrous_coco_11_06_2017"
PATH_TO_CKPT = MODEL_PATH + '/frozen_inference_graph.pb' # frozen model path
PATH_TO_LABELS = os.path.join(MODEL_ROOT, 'object_detection/data', 'mscoco_label_map.pbtxt')
NUM_CLASSES = 90
label_map = label_map_util.load_labelmap(PATH_TO_LABELS)
categories = label_map_util.convert_label_map_to_categories(label_map, max_num_classes=NUM_CLASSES, use_display_name=True)
category_index = label_map_util.create_category_index(categories) # 格式为{1:{'id': 1, 'name': 'person'}, 2: {'id': 2, 'name': 'bicycle'}, ...}
# 模型加载:test.py
detection_graph = tf.Graph()
with detection_graph.as_default():
od_graph_def = tf.GraphDef()
with tf.gfile.GFile(PATH_TO_CKPT, 'rb') as fid:
serialized_graph = fid.read()
od_graph_def.ParseFromString(serialized_graph)
tf.import_graph_def(od_graph_def, name='')
# 防止内存不足,限制sess内存使用比例
gpu_memory_fraction = 0.4
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=gpu_memory_fraction)
config = tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False, allow_soft_placement=True)
config.gpu_options.allow_growth = False
def detect(image_path):
with detection_graph.as_default(): # 需要手动close sess
with tf.Session(graph=detection_graph, config=config) as sess:
image = cv2.imread(image_path)
image_np = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)
image_np_expanded = np.expand_dims(image_np, axis=0)
image_tensor = detection_graph.get_tensor_by_name('image_tensor:0')
boxes = detection_graph.get_tensor_by_name('detection_boxes:0')
scores = detection_graph.get_tensor_by_name('detection_scores:0')
classes = detection_graph.get_tensor_by_name('detection_classes:0')
num_detections = detection_graph.get_tensor_by_name('num_detections:0')
(boxes, scores, classes, num_detections) = sess.run(
[boxes, scores, classes, num_detections],
feed_dict={image_tensor: image_np_expanded})
vis_util.visualize_boxes_and_labels_on_image_array(
image_np,
np.squeeze(boxes),
np.squeeze(classes).astype(np.int32),
np.squeeze(scores),
category_index,
use_normalized_coordinates=True,
line_thickness=4)
new_img = cv2.cvtColor(image_np, cv2.COLOR_RGB2BGR)
cv2.imshow("test", new_img)
cv2.waitKey(0)
if __name__ == '__main__':
detect(/your/test/image)
参考
https://github.com/tensorflow/models/tree/master/object_detection
https://blog.metaflow.fr/tensorflow-how-to-freeze-a-model-and-serve-it-with-a-python-api-d4f3596b3adc
https://www.tensorflow.org/extend/tool_developers/