Skip to content

Latest commit

 

History

History
106 lines (67 loc) · 3.89 KB

README.md

File metadata and controls

106 lines (67 loc) · 3.89 KB
Raw

Carton-box-detector-in-a-video-using-Tensorflow-object-detection-API

This model will detect carton boxes in a video(like from CCTV camera etc),and count them .This is a application in used during supply chain management and logistics.

We use tensorflow object detection API to build a custom object recognition model.

METHOD 1

Pre requisites

Prerequisite packages:

  • pillow
  • lxml
  • jupyter
  • matplotlib
  • opencv

All the above packages can be installed by "pip install command" Before starting we need to install and dowload few files. Refer: https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/install.html#tensorflow-models-installation

LabelImg

Downloading labelImg: follow this link: https://github.com/tzutalin/labelImg After this we will be able to get images with there .xml file that stores the bounding box values

Downloading the TensorFlow Models

Creating TF records:

There are two steps in doing so:

  1. Converting the individual *.xml files to a unified *.csv file for each dataset
  • Create a new file with name xml_to_csv.py

Run the following commands:

Create train data:

python xml_to_csv.py -i [PATH_TO_IMAGES_FOLDER]/train -o [PATH_TO_ANNOTATIONS_FOLDER]/train_labels.csv

Create test data:

python xml_to_csv.py -i [PATH_TO_IMAGES_FOLDER]/test -o [PATH_TO_ANNOTATIONS_FOLDER]/test_labels.csv

  1. Converting the *.csv files of each dataset to *.record files (TFRecord format)
  • Create a new file with name generate_tfrecord.py

Run the following commands:

python generate_tfrecord.py --label= --csv_input=<PATH_TO_ANNOTATIONS_FOLDER>/train_labels.csv --img_path=<PATH_TO_IMAGES_FOLDER>/train --output_path=<PATH_TO_ANNOTATIONS_FOLDER>/train.record

python generate_tfrecord.py --label= --csv_input=<PATH_TO_ANNOTATIONS_FOLDER>/test_labels.csv --img_path=<PATH_TO_IMAGES_FOLDER>/test --output_path=<PATH_TO_ANNOTATIONS_FOLDER>/test.record

To understand how xml package of python works , you can refer:

https://www.datacamp.com/community/tutorials/python-xml-elementtree

Configuring a Training Pipeline

  • The model we shall be using in our examples is the faster_rcnn_inception_v2_coco model.
  • Goto and we can choose our own model based on our performance and speed.
  • Download the .config file.
  • We need to change neccessary parameters to make model effecient.

1.Choose good learning rate.

2.Choose efficient optimizer.

3.Try and change hyperparameters etc

  • Run the following command to start the training: python train.py --logtostderr --train_dir=training/ --pipeline_config_path=training/faster_rcnn_inception_v2_coco.config

  • According to the results we get , we try to trade off with other parameters and hyperparameters.I stoped the model training for 68845 steps for a loss of 0.02**.

Monitor Training Job Progress using TensorBoard

To start a new TensorBoard server, we follow the following steps:

  • type this in your cmd prompt: activate tensorflow_gpu
  • cd into the training_demo folder.
  • Run the following command: *tensorboard --logdir=training*

To visualize the graphs you can go to these url in you browser :http://YOUR-PC:6006 or http://localhost:6006

Finally in order to test the model on live video use webcam_demo.py file (This is a general object detection code) and change model to your model and run the code.

The model has been tested on the_output_final1 ,the_output_final2 ,the_output_final3 video files.

METHOD 2

Reference: https://tensorflow-object-detection-api-tutorial.readthedocs.io/en/latest/