Training-SSD-Object-Detection

The First: You install Tensorflow Object Detection API.

you need to make sure you have completed this step

The Second: Prepare Data and Config

You're download config from ssd_mobilenet_v2_fpnlite_320x320_coco17_tpu-8.tar.gz and copy into models folder.

(I copied, but you must extract file .gz because you're going to use checkpoint file).

Prepare Data

• Relace your data into images and then:
  • Create label map with Create_Label_Map.py and fix labels in cmd with python at Training-SSD.
  • Create TF_Record with each command in Create_TF_Record.txt

Prepare Config file

• Repair #num_classes.
• Repair #batch_size (e.g 1, 2, 4, 8, 16, 32, 64, ...).

The Third: Training model

• run this in cmd:

  python C:/NguyenDatThanh/TF2_ObjDetect_API/models/research/object_detection/model_main_tf2.py --model_dir=models --pipeline_config_path=models/pipeline.config --num_train_steps=5000
  

Note: model_main_tf2.py file depends on your path when you have done step the first.

You can try to change num_train_steps=50000.

• Then you follow link to use model checkpoint