How To Train Object Detection API

I Believed that you've already know what the Object Detection API is.

This tutorial is to provide a fast training script.

1. Install anaconda
2. git clone https://github.com/tensorflow/models.git
3. Create conda virtual environment include tensorflow and python

https://github.com/tensorflow/models/blob/master/research/object_detection/g3doc/installation.md

4. conda install Cython contextlib2 pillow lxml jupyter matplotlib

goto models/research folder

5. Protobuf

# From tensorflow/models/research/
protoc object_detection/protos/*.proto --python_out=.

6. modifly the model_main.py in models

/research/object_detection/model_main.py
@@ -25,6 +25,8 @@ import tensorflow as tf
 from object_detection import model_hparams
 from object_detection import model_lib
 
+tf.logging.set_verbosity(tf.logging.INFO)
+
 flags.DEFINE_string(
     'model_dir', None, 'Path to output model directory '
     'where event and checkpoint files will be written.')
@@ -59,7 +61,7 @@ FLAGS = flags.FLAGS
 def main(unused_argv):
   flags.mark_flag_as_required('model_dir')
   flags.mark_flag_as_required('pipeline_config_path')
-  config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir)
+  config = tf.estimator.RunConfig(model_dir=FLAGS.model_dir, log_step_count_steps=1)

7. modify the MODELS_DIR in MajiangQuickStart.sh

8. Gathering and labeling pictures (we have example in majian_data folder)

9. MajiangQuickStart.sh majian_data This will read images in majian_data/train and majian_data/test, and output training process to training

10. To exporting the inference graph Refers to MajiangFrozen.sh

To train your own data

1. Create your own folder, e.g: "pocker"
2. Gathering and labeling pictures
3. separate the picture into pocker/train and pocker/test
4. list the classes in pocker/predefined_classes.txt

To change the model zoo

1. Get your faverate model from model zoo

2. Untar model, copy to this project, and copy the pipeline.config to template/pipeline.template

3. Modify the num_classes: <tag_number>