Test with Ubuntu 14.04, Tensorflow 1.4.

To Run Model

Follow Installation to prepare environment before running.

Path of Installation: object_detection/g3doc/installation.md

Note

Protobuf should be 2.6 version, you can get version by protoc --version, upgrade Protobuf by website: http://blog.csdn.net/sparkexpert/article/details/73456767 .

To Train Model

Look at my note in chinese for details.

Link：https://pan.baidu.com/s/1i5epSVj

Password：b60y

To Use Own Model

1. cd to my_workspace/object_detection, input: protoc object_detection/protos/*.proto --python_out=.
2. open .bashrc and add export PYTHONPATH=/home/zj/my_workspace/object_detection:$PYTHONPATH
3. cd to my_workspace/object_detection to test installation of own model, input: python object_detection/builders/model_builder_test.py

Load images to detect.

Modify from tensorflow/models/research/object_detection/object_detection_tutorial.ipynb.

Use skimage.io to read image which is faster than PIL.

Load images to detect.

Contain a class of object_detection named DetectImage.

Note: Before using DetectImage, OBJECT_DETECTION_PATH, PATH_TO_CKPT, PATH_TO_LABELS and NUM_CLASSES should be modified.

Load a video or camera to detect.

Note:

1. OpenCV with ffmpeg is needed.
2. OBJECT_DETECTION_PATH should be set in 'object_detection_video.py'
3. Some parameters need to modify, you can find them all in DetectVideo class.

Some parameters need to modify, you can find them at the head of file which labeled as # modify.

Some .config files write by me.

Copy from Tensorflow Object Detection API in tensorflow/models and add my own codes.

Some logs for training.