Pixano: Open-Source Dataset Annotation
pixano.cea.fr · checked 30 Sept 2026
Open-source workspace for investigating computer vision datasets and adding AI-supported labels.
How it plugs in
Signal on Image Annotation Software6.4#5 of 28
Teams can use Pixano, an open-source project available at no cost, to inspect and label computer vision datasets. It handles text, images, video, and varied annotation approaches, with semantic search and model-assisted tools. Access is available through a web interface, Python, Docker, or self-hosting. COCO and other formats can be imported or exported, while development remains active.

Review
In a line
As an open-source dataset exploration and annotation tool, Pixano comes from CEA List for AI work in manufacturing, security, robotics, and transportation.
What it does
Teams can inspect Pixano datasets that combine text, images, and video. Its labeling toolkit covers bounding boxes, polygons, pixel masks, keypoints, cuboids, classification, and tracking.
Smart features can help with labels. Models such as CLIP can power semantic search, and SAM can support smart segmentation. Video annotations may propagate across frames, while geometric transformations can align 3D cuboids with point clouds.
How it connects
The application and its datasets are accessible through Pixano's Python and REST APIs. For deployed models, Pixano Inference supplies a Ray Serve-based server together with Python and REST clients.
It uses the Lance format to store and navigate datasets. Import and export include COCO and Pixano's own annotation format; image inputs include JPEG and PNG. Reusable Web Components can be assembled into a custom annotation app.
The project documents pip installation inside a Python virtual environment as well as use of an official Docker image. Supported environments include Linux, macOS, Windows, the web, and self-hosted deployments.
Good fit
Pixano suits AI developers and teams that need to inspect or label computer vision data, including projects involving manufacturing, security, robotics, or transportation.
It is a match when you need several annotation types, model-assisted labeling, semantic search, or a custom interface built from reusable annotation parts.
Watch for
Development remains active, so API behavior can change. The documented Python range starts at 3.10 and stops before 3.14; check your environment before installing.
The official information describes 3D point-cloud support as planned, even though 3D cuboid annotation and point-cloud matching are listed. Check that your required 3D workflow is covered.
The reviewed pages do not name any security certification or compliance standard.
Pricing
Pixano is free. A free plan is available, with no separate trial period.
Verdict
Choose Pixano if you want open-source dataset annotation with APIs, local or self-hosted deployment, and AI-assisted labeling. Its support for COCO, common image files, video, and multiple annotation types can fit varied computer vision projects.
Look elsewhere if you need a fixed API, stated compliance certifications, or confirmed production-ready point-cloud support.
Our take
For teams seeking a no-cost, open-source workspace for computer vision data, Pixano is a sensible option. It combines numerous annotation types with semantic search, model-assisted labeling, APIs, and self-hosted deployment. The main trade-offs are its changing API, specific Python version range, and unclear status for full point-cloud workflows. Verify those details and your security requirements before adopting it for a long-lived or regulated project.
- Free and open-source under the CeCILL-C license
- Supports images, video, text, and many annotation types
- Offers Python and REST APIs
- Can exchange COCO and Pixano formats
- Runs via pip, Docker, web, or self-hosting
- Active development means the API may change
- The documented Python range is 3.10 or later but below 3.14
- Complete 3D point-cloud capability is still described as planned
- No security certification or compliance standard is stated on the reviewed pages
- Pick it ifComputer vision teams labeling mixed datasets; Work that benefits from APIs or self-hosted deployment; AI-assisted image and video annotation
- Skip it ifYou require a stable API with no planned breaking changes; You need certified compliance documentation; Your workflow depends on confirmed point-cloud support
Pricing
Pixano costs nothing. The free plan does not include a separate trial period.
Questions
Is Pixano free?
Yes. Pixano has a free plan and no separate free trial.
What data can Pixano handle?
It supports datasets with text, images, and videos. JPEG and PNG images are listed, while 3D point-cloud support is described as planned.
Which annotation types are available?
Available options include bounding boxes, polygons, pixel masks, keypoints, cuboids, classification, tracking, customizable labels, and temporal propagation for video annotations.
Does Pixano have an API?
Yes. It documents both Python and REST APIs. Pixano Inference also provides clients for models deployed through its inference server.
Can it import and export COCO data?
Yes. COCO is listed among its import and export formats, alongside Pixano's own annotation format.
Where can Pixano run?
Linux, macOS, Windows, the web, and self-hosted setups are supported. The documented installation paths use pip or an official Docker image.