/OD_setup

Primary LanguageJupyter Notebook

TensorFlow 2 Object Detection

Download gitignore using curl

curl https://raw.githubusercontent.com/c17hawke/FSDS-DVC-NLP-Project-with-docs/main/.gitignore > .gitignore

Download init_setup.sh using curl

curl https://raw.githubusercontent.com/c17hawke/general_template/main/init_setup.sh > init_setup.sh

Run init_setup.sh

bash init_setup.sh

tensorflow verification

python -c "import tensorflow as tf;print(tf.config.list_physical_devices('GPU'))"

Installation of Object Detection API

create a TensorFlow directory

mkdir TensorFlow && cd TensorFlow

Clone the TensorFlow models folder here

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

remove .git dir of models repository to avoid git conflicts

add models folder to .gitignore

echo "TensorFlow/models" >> .gitignore

Protobuff Installation/Compilation

  • Visit the link - https://github.com/protocolbuffers/protobuf/releases
  • windows user -
    • search for - protoc-3.20.1-win64.zip
  • for mac users -
  • search for - protoc-3.20.1-osx-x86_64.zip
  • for linux users - 
    sudo apt install -y protobuf-compiler
  • Unzip into root folder and add <PATH TO protoc folder>/bin into system environment variables

Check protobuff version

  • run the following command
cd TensorFlow/models/research
protoc --version

Convert all protobuff files into python files

  • run the following command
cd TensorFlow/models/research
protoc object_detection/protos/*.proto --python_out=.

Install COCO API

pip install git+https://github.com/philferriere/cocoapi.git#subdirectory=PythonAPI

Install Object Detection API

  • From within TensorFlow/models/research/
cp object_detection/packages/tf2/setup.py .
python -m pip install .

test your installation

  • From within TensorFlow/models/research/
python object_detection/builders/model_builder_tf2_test.py

Run examples -

  • Create workspace/example_1 directory in project root

    mkdir -p workspace/example_1

  • cd to workspace/example_1

    cd workspace/example_1

  • Download notebook

    curl https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/_downloads/55b1ed8e083cbc9ca3bfc1c18eb6b860/plot_object_detection_saved_model.ipynb > plot_object_detection_saved_model.ipynb

Custom model training

mkdir workspace/training_demo
cd workspace/training_demo
mkdir -p annotations exported-models models pre-trained-models images/test images/train

Create label map file in training_demo/annotations with name label_map.pbtxt

and write content as -

item {
    id: 1
    name: 'helmet'
}

item {
    id: 2
    name: 'head'
}

item {
    id: 3
    name: 'person'
}

Create TensorFlow Records

  • It is time to convert our annotations into the so called TFRecord format.

curl the generate_tfrecord.py file under root of training_demo dir

curl https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/_downloads/da4babe668a8afb093cc7776d7e630f3/generate_tfrecord.py > generate_tfrecord.py

generate_tfrecord

for train data:

python generate_tfrecord.py -x images/train -l annotations/label_map.pbtxt -o annotations/train.record

for test data:

python generate_tfrecord.py -x images/test -l annotations/label_map.pbtxt -o annotations/test.record

Configuring a Training Job

Go to model Zoo and download SSD ResNet50 V1 FPN 640X640 (RetinaNet50)

  • extract the downloaded model into training_demo/pre-trained-model directory

Configure training pipeline

  • create a folder my_ssd_resnet50_v1_fpm in training_demo/models folder,
  • copy pipeline.config from to my_ssd_resnet50_v1_fpm from pre-trained_model directory,
  • update it as per the documentation - link

Copy training file from TensorFlow/models/reserch/object_detection/ to the root of training_demo folder

cp ../../TensorFlow/models/reserch/object_detection/model_main_tf2.py .

Start training by running the following command -

python model_main_tf2.py --model_dir=models/my_ssd_resnet50_v1_fpn --pipeline_config_path=models/my_ssd_resnet50_v1_fpn/pipeline.config

Exporting a Trained Model

cp ../../TensorFlow/models/research/object_detection/exporter_main_v2.py .

python exporter_main_v2.py --input_type image_tensor --pipeline_config_path ./models/my_ssd_resnet50_v1_fpn/pipeline.config --trained_checkpoint_dir ./models/my_ssd_resnet50_v1_fpn/ --output_directory ./exported-models/my_model