TFOD-Automatic 1.4

Automated Object detection for Beginner using python and Tensorflow

TFOD_Automatic

Tensorflow object detection automatically from user input.

TFOD_Automatic Setup

In this we going to setup requirements for object detection automatically using TFOD_Automatic.setup.

from TFOD_Automatic import setup

start_download()

start_download is used to download the tensorflow model garden from the github.If the model garden is present already in the working directory.it automatically skip this process.

from TFOD_Automatic.setup import start_download
start_download()

check_model()

check_model is used to check whether the model garden folder is present or not.if the folder is present it return True and else it return False.

from TFOD_Automatic.setup import check_model
check_model()

protobuf_setup()

protobuf_setup is used to setup the protobuf.Protobufs to configure model and training parameters.The Protobuf version is 3.19.

Steps: - Protobuf is downloaded successfully and saved in folder protobuf. - Now copy the folder protobuf folder and Go to C:/Program File and paste it inside - Now copy the location of the file like “C:/Program File /protobuf/bin” - Now open Edit envirnoment variable and go to Enviroment Variable and paste the location inside Path and save. - Finaly the open new Anaconda comment prompt and start project.

To Check whether the protobuf is set or not use check_protobuf() method.

from TFOD_Automatic.setup import protobuf_setup
protobuf_setup()

check_protobuf()

check_protobuf is used to check whether the protobuf is setup is completed or not.Returns true if the setup is correct and False if the setup is incorrect.

from TFOD_Automatic.setup import check_protobuf
check_protobuf()

Note: - Now the tensorflow object detection setup is completed.Next we are going to install the require package automaticaly

TFOD_Automatic Install Package

In this we going to install the require package for tensorflow object detection automatically using TFOD_Automatic.install.

from TFOD_Automatic import install

install_packages()

install_packages is used to install the required Packages like cython,cocoapi,labelImg,pycocotools-windows.To install these package we need visual studio c++ build support.If the visual studio c++ build support is not install then first install it then only we can install the package. All require package get installed.

from TFOD_Automatic.install import install_packages
install_packages()

check_installed()

Check the all packages are installed or not.If the require package is not install it show the error.

from TFOD_Automatic.install import check_installed
check_installed()

install_all()

install_packages is used to install the required Packages like cython,cocoapi,labelImg,pycocotools-windows.To install these package we need visual studio c++ build support.If the visual studio c++ build support is not install then first install it then only we can install the package. All require package get installed and check all the packages are installed.

from TFOD_Automatic.install import install_all
install_all()

Note:

• Now the tensorflow object detection setup is completed and every package is installed.Now we are going to create the object detection model.

To create the tensorflow object detection model there are two method in TFOD_Automatic they are CreateModel and BuildModel:

• BuildModel: It is beginner-friendly.Example one command is used to create model and complete the training.

• CreateModel: It is some advance method. It has many command to create ,train, export the model.

BuildModel:

Download The Tensorflow model garden and Build object Detect model automaticaly.

from TFOD_Automatic import model

ModelSetup():

ModelSetup class is used to automatically create the Folder setup in working directory.And also check every package are installed and setup is completed or not.

from TFOD_Automatic.model import ModelSetup
model_setup=ModelSetup()

BuildModel():

BuildModel is used to automatical create the model.BuildModel complete Generate the data file,Download the pretrained model,Create Pipeline file,export the model.

Generate the data file:

It generate record file from the input image.

Download the pretrained model:

It download the model from the tensorflow model zoo.It is pretrained model.

Create Pipeline file:

Pipeline file contains of the number of training steps,batch size,train and test folder path also the record file path etc.

Export the model:

Finally we need to export the model for future use.

Parameter:

• number_of_class:int default:NO DEFAULT VALUE Number of training set

• pretrained_model_name:string default:ssd_mobilenet_v2_320x320_coco17_tpu Please copy pretrained model name from the tensorflow model zoo

• modelUrl:string default:http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz Pretrained model is downloaded from here.Copy url from tensorflow model zoo

• batch_size_for_train:int default:8 batch while training the model

• train_steps:int default:2000 Training epochs

from TFOD_Automatic.model import BuildModel
build_model=BuildModel(number_of_class)

Notes

• It creates the model with number_of_class with batch_size of 8 and train_step of 2000.

• It takes some time. so be patient.

• After the training is completed their will be the folder called ->workspace->training_demo->exported-models. You will find the exported model.It can used for future use.

CreateModel

• number_of_class:int default:NO DEFAULT VALUE Number of training set

• pretrained_model_name:string default:ssd_mobilenet_v2_320x320_coco17_tpu Please copy pretrained model name from the tensorflow model zoo

• modelUrl:string default:http://download.tensorflow.org/models/object_detection/tf2/20200711/ssd_mobilenet_v2_320x320_coco17_tpu-8.tar.gz Pretrained model is downloaded from here.Copy url from tensorflow model zoo

• batch_size_for_train:int default:8 batch while training the model

• train_steps:int default:2000 Training epochs

from TFOD_Automatic.model import CreateModel
# create the class instance
createmodel=CreateModel()

# model_setup() is used to check whether the package is installed and setup is correct completed.
createmodel.model_setup()

# create_folder() is used to create the require folder in the working directory
createmodel.create_folder()

# image_data() is used to generate the label map in .pbtxt format
createmodel.image_data()

# generate_file() is used to generate the .record file for the train and test image
createmodel.generate_file()

# download_model() is used to Download the pretrained model from tensorflow model zoo.
createmodel.download_model()

# configure_pipeline() used to create Pipeline file:Pipeline file contains of the number of training steps,batch size, train and test folder path also the record file path etc.
createmodel.configure_pipeline()

#export_model() is used to export the model.Finally we need to export the model for future use.
createmodel.export_model()