Preparation

0. create a follder<-path_1
   
1. download everything from the current project to path_1
   Note that all the prerequisites are from tensorflow official models
   
2. copy official_model/research/object_detection to path_1
   
3. copy official_model/research/object_detection/utils&models to path_1
   
4. copy official_model/research/slim/datasets&deployment&nets to path_1
   
5. copy official_model/research/cvt_text/model to path_1
   
6. download link for ssd_model and extract it to path_1
   
7. install necessary py packages
   

Fix data (from current project)

1. object_Dataset/images contains source images
   
2. object_Dataset/annotations contains annotations information for source images
   
3. if you would like to annotate images by yourself, use the label_tool
   
4. object_Dataset/train.txt contains image names for training
   
5. object_Dataset/test.txt contains image names for testing
   
6. object_Dataset/object_label_map.pbtxt contains label information
   
7. set correct paths in the config_modification.py and generate the config.xml
   
8. set correct paths of "fine_tune_checkpoint","train_input_reader","eval_input_reader" in /ssd_mobilenet_v1_small_object.config
   

Train

1. run /totfrecord.py twice (path configuration needs to be motified) to generate tfrecords for training and testing respectively, check config.xml if error occurs
   
2. run /train.py to train you model, press "control_c" to stop when receiving a stable loss
   

Freeze model

1. run /export_inference_graph.py to freeze the *.pb model, output is supposed to be saved at the train folder (latest step by default)
   

test

1. set path configuration in /object_detection_my.py and run