DevMesh

Localizing birds from pictures using TensorFlow Object Detection API and OpenVINO

Sayak Paul

Sayak Paul

Kolkata, West Bengal

0 0
  • 0 Collaborators

This project demonstrates the use of the TensorFlow Object Detection API along with OpenVINO to localize birds from a given image. ...learn more

Project status: Under Development

Artificial Intelligence

Intel Technologies
OpenVINO

Overview / Usage

This project can help researchers to identify different birds from still images or live video feeds. The project is being developed using the Caltech-UCSD Birds-200-2011 dataset which consists of 200 different bird categories.

Methodology / Approach

I used the TensorFlow Object Detection API to train a custom object detection model using the above-mentioned dataset. I fine-tuned the SSDMobileNetV1 model which was pretrained on the Coco dataset. I trained the model for 100k steps and I got an **mAP **of ~ 77.5%. For the purpose of this project, I reduced the number of classes to 20.

After generating the frozen inference graph, I used OpenVINO to further optimize it for running inference on an NCS.

Technologies Used

  • TensorFlow Object Detection API

  • GCP's ML Engine to faciliate multi-gpu training

  • OpenVINO to optimize network inference

Comments (0)