EfficientVis: Explore EfficientNet Models with Ease

EfficientVis, is a simple website for exploring the capabilities of EfficientNet models pre-trained on ImageNet-1k. Designed with simplicity in mind, EfficientVis offers a user-friendly platform where anyone can effortlessly test all of the available EfficientNet architectures on their own images.

Key Features

• Broad Model Coverage: Experiment with the entire range of pre-trained EfficientNet and EfficientNetv2 models. From EfficientNet-B0 to EfficientNet-B7, as well as all three versions of EfficientNetV2 explore the spectrum of capabilities and differences between the models.

• Custom Prediction Counts: Specify the number of class predictions you would like to be displayed, find all ImageNet classes here.

• Image Transformation Insights: Understand exactly what the model sees by observing the modifications your images undergo before prediction.

• Normalised Predictions: EfficientVis offers a normalisation option of only the predictions shown to help you derive more indicative results.

• Local, Fast, and Secure: EfficientVis is designed to run locally, ensuring that your efficient and secure interactions.

Getting Started

API Setup:

1. Install all required Python packages by running the following command:

   pip install -r /flask-api/requirements.txt

2. Then start the API within the flask-api folder by executing:

   python app.py

Website Setup:

1. Ensure Node.js and npm are installed on your system.

2. While in the front-end folder, run the following commands to install dependencies and start the development server:

   npm install
   npm run dev

Quick Setup

For Unix systems, the setup_project.sh script can be used for quick setup:

In the root project directory:

1. Ensure the shell script is executable by running:

   chmod +x setup_project.sh

2. Execute the script to set up the project:

   ./setup_project.sh

Note:

• When predicting wiith a model for the first time, models will be downloaded to /Users/user/.cache/torch/hub/checkpoints/model_name.pth from https://download.pytorch.org/models/model_name.pth.
• Models vary in size, with the largest being 454.6 MB. You can check each for each one here and here.

What's Next

In my desire to learn the intricacies of deep learning and visualise the magic behind neural nets, my vision is to gradually make a scalable platform where people can test and explore the capabilities of different deep learning models remotely, in computer vision and beyond.

Acknowledgments

This project utilizes the EfficientNet model through its PyTorch implementation by Luke Melas-Kyriazi, based on the original EfficientNet models developed by Mingxing Tan and Quoc V. Le. A big thanks to the authors for their significant contributions to the field of deep learning.

Model and Implementation

For detailed information on the model and its PyTorch implementation, visit the EfficientNet-PyTorch GitHub repository.

License

This project is built upon work licensed under the Apache License 2.0. The original license can be found here.

Your support and contributions to this project are greatly appreciated. If you find any bugs or have suggestions for improvement I would love to hear from you!