Tensorflow-Object-Detection-API

This is a Custom Object Detection using TensorFlow from Scratch. Here I have done a multi object classification for Kitchen Utilities.

You can find the detailed blog about this in this blog. Link

You can also download the dataset from the link metioned below.

Link: http://homepages.inf.ed.ac.uk/rbf/UTENSILS/

Downloading Git repos

Download the full TensorFlow object detection repository located at https://github.com/tensorflow/models by clicking the “Clone or Download” button and downloading the zip file.

Along with this you need to download the Tensorflow Model git repo and faster rcnn model from TensorFlow's model zoo

Faster-RCNN-Inception-V2 model. Download the model here.

Supporting files and models

The Python modules files that supports for this project is as shown below,

C:\> pip install protobuf
C:\> pip install pillow
C:\> pip install lxml
C:\> pip install matplotlib
C:\> pip install numpy
C:\> pip install scipy
C:\> pip install sklearn
C:\> pip install pandas
C:\> pip install opencv-python

## Make sure that your always in admin mode when you work with protoc, if not you defenitely encounter error

You may encounter an error which is No module named 'deployment' or No module named 'nets'

to solve this do the following,

set PYTHONPATH=C:\tensorflow2\models
set PYTHONPATH=C:\tensorflow2\models\research
set PYTHONPATH=C:\tensorflow2\models\research\slim

Once your done with this, you need to change the directory and then execute the following code

C:\tensorflow2\models\research>
C:\tensorflow2\models\research> python setup.py build
C:\tensorflow2\models\research> python setup.py install

LabelImg

LabelImg is a great tool for labeling images, and its GitHub page has very clear instructions on how to install and use it.

LabelImg GitHub link

LabelImg download link

or you also install with the help of python

C:\> pip install labelimg
C:\> labelimg

finally you will be having the following output

Generating TFRecord code

Make sure your inside the folder

C:\tensorflow2\models\research\object_detection

python generate_tfrecord.py --csv_input=images\train_labels.csv --image_dir=images\train --output_path=train.record
python generate_tfrecord.py --csv_input=images\test_labels.csv --image_dir=images\test --output_path=test.record

Execute the Training Program

Make sure your inside the folder, and execute the following code

C:\tensorflow2\models\research\object_detection

python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/faster_rcnn_inception_v2_pets.config