🚀 Pipeless Agents is now live! Create vision powered automations effortlessly within seconds. Click here

Pipeless

The open-source computer vision framework

Enjoy the best development experience analyzing and manipulating video in real-time

Easily integrate with industry standard models and libraries

Start your project

Documentation

Pipeless is part of Nvidia Inception

Function-oriented development experience


        def hook(frame_data, context):
            rgb_frame = frame_data['original']

            reduced_frame = cv2.resize(
                rgb_frame, (reduced_width, reduced_height))

            frame_data['inference_input'] = np.array(
                reduced_frame
            )

pre-process.py


        {
            "runtime": "onnx",
            "model_uri": "https://downloadable_model_file",
            "inference_params": {
                "execution_provider": "TensorRT"
            }
        }

process.json


        def hook(frame_data, _):
            frame = frame_data['original']
            b_boxes = frame_data['inference_output']

            for box in b_boxes:
                a, b, w, h = resize_bbox(box)
                cv2.rectangle(
                    frame, (a, b),
                    (a + w, b + h), (255, 0, 255), 2
                )

            frame_data['modified'] = frame

post-process.py

What is Pipeless?

Pipeless is a framework for developers to create and deploy computer vision applications in just minutes. It takes care of everything you need around a model to put your application in production. That includes code parallelization, multimedia pipelines, memory management, model inference, multi-stream management, and more. You simply write self-contained functions that take a frame as input and Pipeless takes care of running them when required.
Pipeless runs anywhere, you can deploy your applications to the cloud or to edge devices, even without an internet connection.
You can fully automate your application workflows by performing actions through a CLI or REST API, such as adding, updating or removing streams.

Multi-stream processing is provided out-of-the-box, even using different inputs and outputs with protocols such as files, RTSP, RTMP, HTTP, etc.
You can also provide restart policies to your streams, so they automatically restart the processing when the stream reaches the end, when there is a camera failure, etc.

Pipeless is everything you need to develop scalable computer vision applications.

Pipeless License

Integrate Pipeless in your commercial product.

Get the enterprise version with enhanced features, better performance and ongoing maintenance.

$30 per camera/stream monthly.

Let's talk 15 min!

Get started pack

Perfect for evaluating Pipeless and integrate with your product.

Work 10 hours with our team on your product!

Join the community

Supported by a network of early advocates and contributors

GitHub discussions

Discord

Subscribe to our newsletter!

Receive new tutorials, community highlights, release notes, and more!

Example tutorials

Pose Estimation

With pose estimation you can identify and track the positions of key body joints. It enables applications such as gesture recognition, human-computer interaction, exercise monitoring, biomechanical analysis and others. It is widely used in fields like gaming, healthcare, and robotics.

See step by step tutorial

Object detection and tracking

With object detection you can locate and identify elements. It plays a crucial role in various applications, including autonomous vehicles, surveillance, and image-based search, among others. You can identify and track multiple objects simultaneously.

See step by step tutorial

Pipeless Features

Focus on what matters

Create and deploy your use case in minutes. Forget about building and maintaining multimedia pipelines.

Multi-stream support

Pipeless allows you to process any number of streams at the same time in a lightweight way.

Dynamic stream management

Add, edit or remove streams dynamically with just a command or HTTP request.

Automatic inference

Pipeless integrates some of the widely adopted inference runtimes, such as the ONNX Runtime, OpenVINO, CoreML, CUDA, TensorRT, etc.

Lightning speed - Real Time

Pipeless parallelizes the execution for you so you don't have to think about it. Run your use case in real-time without effort.

Multi-language support

Write your application code in many languages, even mixing them on a single application. Python, Rust, ... it doesn't matter.

Low Code

Effortlessly craft your app in minutes by reusing black boxes, or delve deeper and write your custom code for a truly customized experience

Developer Focused

Leverage the development experience of serverless functions but for embedded computer vision. Developed by and for developers. We understand your needs.

Edge, IoT or Cloud

Whether you app needs to run embedeed on edge or IoT devices or if it runs in the cloud, we have you covered.

Open Source

Pipeless is an Open Source project. You can use it, customize it and run it on your own infrastructure if you desire. Forget about vendor lock-in.

BLOG

Pipeless

Offload