Frugally Deep: Keras Inference in C++

Understand the project

Practical overview

When it fits

Use it when model training can remain in Keras/Python but prediction needs to run in a C++ application without linking that application against TensorFlow, and the model's layers and configuration are supported.

Who it is for

C++ developers integrating inference from a Keras model into an application or pipeline.
Teams building models in Keras/Python that want to run predictions in C++.
Developers evaluating a smaller-footprint alternative to linking an application against TensorFlow.

Documented uses

Convert a saved Keras model into Frugally Deep's file format.
Load the converted model from C++ and pass input tensors to its prediction function.
Integrate supported sequential or functional-API model graphs into a C++ program.

Before you choose

Requirements & limits

The model must use the TensorFlow backend and channels-last image data format.
Layer support is broad but not universal; Lambda, text and spectrogram layers, stateful recurrent layers and temporal models are among the listed unsupported cases.
The library depends on FunctionalPlus, Eigen and nlohmann/json, which are also header-only.
Each prediction uses one CPU core and the library does not use a GPU; parallel predictions can use additional CPU cores.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Convert a Keras model for C++ inference

Prepare a Keras model file and load the converted model for prediction from C++.

Before you start

A Keras model using the TensorFlow backend and channels-last image data format.
Python/Keras for saving the model and running the conversion script.
A C++ environment with Frugally Deep and its listed dependencies: FunctionalPlus, Eigen and nlohmann/json.

Steps

Build and train the model in Keras/Python, then save it as a .keras file.
Run the command below with the saved model file and a destination JSON filename.
Documented command
```
python3 keras_export/convert_model.py
```
In C++, include `, load the JSON file with fdeep::load_model(...), and call model.predict(...)` with input tensors.

Expected outcome

A converted model can be loaded in C++ and used for a forward-pass prediction; loading also runs the generated comparison test case.

Guide limitations

Only supported model layers and operations can be used; the README lists Lambda layers, several text and spectrogram layers, stateful recurrent layers and temporal models as unsupported.
Models using a different image data format or a backend other than TensorFlow are not supported.
Prediction uses one CPU core and does not use a GPU.

Documentation 1 ↗

COMPARE OPTIONS

Compare options for Model inference and serving

These projects document a common task and share a project role. This is a source-based comparison, not an execution benchmark or a guarantee of interchangeability.

Project	Documented purpose	Platform	Access and requirements
Frugally Deep	Convert a saved Keras model into Frugally Deep's file format.	Linux, macOS, Self-hosted, Windows
Hyperpose	Customize pose-estimation architectures and datasets.	Docker, Linux, macOS, Windows
Multi Task Nlp	Define multiple tasks and their data and model settings in a YAML file.	Linux, macOS, Windows	The project is an open-source library under the Apache-2.0 license, distributed through its public GitHub repository.

Evidence and transparency

Sources & review

Official repository ↗

Source record details

Source availability: Active
Sources checked: 13 October 2026
Original website title: Dobiasd/frugally-deep · Official homepage title · retained in its source language

How we review sources Suggest a correction

Frugally Deep