If you’re looking to harness the power of TensorFlow for object detection without the need for a GPU, you’re in the right place! This guide will walk you through the steps necessary to set up the TensorFlow CPU Inference API on both Windows and Linux systems. Moreover, we will explore how to extract text from images using optical character recognition (OCR).

Prerequisites

• Operating System:
  • Ubuntu 16.04 or 18.04
  • Windows 10 Pro/Enterprise
• Docker installed

Checking for Prerequisites

To verify if you have Docker installed, run the following command:

docker --version

Installing Prerequisites

For Ubuntu

Run the following commands to install Docker:

chmod +x install_prerequisites.sh && source install_prerequisites.sh

For Windows 10

You can install Docker by following the instructions at this link.

**P.S: After installation, open Docker Desktop and adjust resources under Settings and Advanced**

Building the Docker Image

To create the project’s Docker image, execute the following command from the project’s root directory:

sudo docker build -t tensorflow_inference_api_cpu -f docker/Dockerfile .

Behind a Proxy?

If you’re behind a proxy, use the command like this:

sudo docker build --build-arg http_proxy= --build-arg https_proxy= -t tensorflow_inference_api_cpu -f docker/Dockerfile .

Running the Docker Container

You can deploy this API using either **docker** or **docker swarm**. If there’s no specific requirement for redundancy or scaling, just use docker. To run the container, navigate to the API’s directory and execute:

Using Linux-based Docker:

sudo docker run -it -v $(pwd)/models:models -v $(pwd)/models_hash:models_hash -p :4343 tensorflow_inference_api_cpu

Using Windows-based Docker:

docker run -it -v $PWD/models:models -v $PWD/models_hash:models_hash -p :4343 tensorflow_inference_api_cpu

The docker_host_port can be any unique port of your choice. Once executed, the API will be ready to accept HTTP requests on that port.

Understanding the API Endpoints

After running the API, access the available endpoints by navigating to: http://:/docs. Here’s a summary of important endpoints you can use:

• load (GET) – Loads all available models.
• detect (POST) – Performs inference on a specified model and returns bounding boxes.
• get_labels (POST) – Returns all the specified model labels.
• models/model_name/ocr (POST) – Uses images to extract text details.

**Note:** For custom endpoints like load, detect, and get_labels, always use the load endpoint first.

Model Structure

The folder named models contains subfolders for each model. Each subfolder must include:

• pb file — frozen_inference_graph.pb (model weights).
• pbtxt file — object-detection.pbtxt (model classes).
• Config.json — Configuration information for the model.

Troubleshooting

If you encounter issues during deployment or API usage, consider the following:

• Ensure Docker is correctly installed and running.
• Check if the specified Docker host port is not in use by another application.
• Review the API endpoint calls for correct syntax.

For more insights, updates, or to collaborate on AI development projects, stay connected with fxis.ai.

At fxis.ai, we believe that such advancements are crucial for the future of AI, as they enable more comprehensive and effective solutions. Our team is continually exploring new methodologies to push the envelope in artificial intelligence, ensuring that our clients benefit from the latest technological innovations.