Getting Started

Setup Virtual Environment

python -m venv env

source env/bin/activate # Linux

.\env\Scripts\activate # Windows

deactivate # exit from virtual environment

Optional if use jupyter

pip install ipykernel jupyter

python -m ipykernel install --user --name=env

LABELIMG

pip install opencv-python

collect image, store to images folder, 20% for folder test and 80% for folder train in images folder

pip install --upgrade pyqt5 lxml

git clone https://github.com/tzutalin/labelImg

cd labelImg && pyrcc5 -o libs/resources.py resources.qrc

cd labelImg && python labelImg.py (label the images from folder train n test)

SETUP PROJECT

mkdir Project

copy images folder after labelImg to Project

cd Project

git clone https://github.com/tensorflow/models.git

download n extract protoc https://github.com/protocolbuffers/protobuf/releases/download/v3.15.6/protoc-3.15.6-win64.zip

copy protoc.exe to models/research

INSTALLATIONS

cd models\research

Compile protos.

protoc object_detection/protos/*.proto --python_out=.

Install TensorFlow Object Detection API.

cp object_detection/packages/tf2/setup.py .

python -m pip install .

Test the installation.

python object_detection/builders/model_builder_tf2_test.py #Ran 24 tests in 28.539s OK (skipped=1), if not like this, install the missing module base on console

python and type import object_detection then enter

*if error, install the missing module base on console

GENERATE TFRECORD

cd to root project, mkdir data, create label_map.pbtxt in data folder

adjust label_map.pbtxt with label name such as => item { name:'car' id:1 } item { name:'handphone' id:2 }

cd to root project, git clone https://github.com/nicknochnack/GenerateTFRecord

python GenerateTFRecord\generate_tfrecord.py -x images\train -l data\label_map.pbtxt -o data\train.record

python GenerateTFRecord\generate_tfrecord.py -x images\test -l data\label_map.pbtxt -o data\test.record

TRAIN MODEL

download and extract zoo model in SSD MobileNet V2 FPNLite 320x320

mkdir mymodels

copy pipeline.config in zoo model to mymodels

Update Config For Transfer Learning

adjust pipline.config in mymodels (num_class, batch_size, fine_tune_checkpoint, fine_tune_checkpoint_type="detection" label_map_path, input_path)

or :

adjust config in script\update-pipline-config.py then run python script\update-pipline-config.py

Start Training

python models\research\object_detection\model_main_tf2.py --model_dir=mymodels --pipeline_config_path=mymodels\pipeline.config --num_train_steps=2000

EVALUATE MODEL

python models\research\object_detection\model_main_tf2.py --model_dir=mymodels --pipeline_config_path=mymodels\pipeline.config --checkpoint_dir=mymodels

cd mymodels\train && tensorboard --logdir=.

ISSUE in opencv-python

pip uninstall opencv-python-headless -y

pip uninstall opencv-python -y

pip install opencv-python

--> TEST DETECTION with Load Train Model From Checkpoint <--

python script\ckpt-detect-img-cv2.py

python script\ckpt-detect-video.py

FREEZING THE GRAPH

cd mymodels && mkdir export

then, in root projcet run:

python models\research\object_detection\exporter_main_v2.py --input_type=image_tensor --pipeline_config_path=mymodels\pipeline.config --trained_checkpoint_dir=mymodels --output_directory=mymodels\export

EXPORT INFERENCE_GRAPH FOR Tensorflow 1.X

python models\research\object_detection\export_inference_graph.py --input_type image_tensor --pipeline_config_path mymodels\pipeline.config --trained_checkpoint_prefix mymodels\model.ckpt-3 --output_directory mymodels\inference_graph

*note adjust model.ckpt-{highestCount}

*node this script not work in TF2

EXPORT TO TFJS USE FOR WEB APP

pip install tensorflowjs

cd mymodels && mkdir tfjsexport

then, in root projcet run:

tensorflowjs_converter --input_format=tf_saved_model --output_node_names='detection_boxes,detection_classes,detection_features,detection_multiclass_scores,detection_scores,num_detections,raw_detection_boxes,raw_detection_scores' --output_format=tfjs_graph_model --signature_name=serving_default mymodels\export\saved_model mymodels\tfjsexport

CONVERSION TO TFLITE USE FOR MOBILE OR RASPY

cd mymodels && mkdir tfliteexport

then, in root projcet run:

python models\research\object_detection\export_tflite_graph_tf2.py --pipeline_config_path=mymodels\pipeline.config --trained_checkpoint_dir=mymodels --output_directory=mymodels\tfliteexport

tflite_convert --saved_model_dir=mymodels\tfliteexport\saved_model --output_file=mymodels\tfliteexport\saved_model\detect.tflite --input_shapes=1,300,300,3 --input_arrays=normalized_input_image_tensor --output_arrays='TFLite_Detection_PostProcess','TFLite_Detection_PostProcess:1','TFLite_Detection_PostProcess:2','TFLite_Detection_PostProcess:3' --inference_type=FLOAT --allow_custom_ops

--> TEST DETECTION FROM EXPORTED MODEL <--

python script\image-detection.py # for image detection

python script\realtime-detection.py # for realtime (webcam or video) detection

Hope you enjoy it! 😁