We have been annotating data across medical, geospatial, automotive, and industrial projects for nine years. In that time, we have tested LiDAR annotation tools and learned that the right one depends entirely on what you are annotating and what you need the output to do. Some tools that look impressive in a demo are painful in production. Some simple tools that nobody writes about do exactly what they need to do, reliably, without overhead. with a lot of annotation tools and learned that the right one depends entirely on what you are annotating and what you need the output to do. Some tools that look impressive in a demo are painful in production. Some simple tools that nobody writes about do exactly what they need to do, reliably, without overhead.
This is the third article in our Tools We Love series 2026 edition. Episode 1 covered thThe Best Computer Vision Annotation Tools in 2026 and episode 2 covered the Best Enterprise Data Annotation Platforms in 2026.
How We Evaluated LiDAR annotation tools & Open Source Annotation Tools
Our reviews draw from HITL ops and tech teams, who work with these platforms in different capacities. Yalda, our Project Manager, evaluates tools from a client-facing standpoint. Sai, our Tech Coordinator, assesses infrastructure requirements, integration depth, and ML pipeline fit. Joy, project coordinator and Mercy, our QA specialist evaluate annotation output quality, review workflows, and what breaks under production conditions.
Every tool reviewed here has been used on real projects, with real data, by the HITL team.
VGG Image Annotator (VIA) - The Zero-Setup Research Annotation Tool
VGG Image Annotator , VIA, was built by the Visual Geometry Group at Oxford University and released under the BSD-2 clause license. It runs entirely in a web browser as a single HTML file under 500 kilobytes. No installation, no server, no account. You download the file, open it in a browser, and start annotating. That simplicity is the point.
Mercy, our QA Specialist, recommends it specifically when simplicity and zero cost are the priority: “It’s simple and easy to use. It’s open source, which makes it more attractive.” The annotation type support covers the essentials, bounding boxes, polygons, circles, ellipses, points, and polylines, which handles the majority of region annotation tasks that researchers and solo annotators encounter. Export goes to CSV and JSON, which feeds cleanly into TensorFlow, PyTorch, and YOLO pipelines without conversion overhead.
The limitations are real and worth stating clearly before committing to it. For example, VIA has no built-in team management, user roles, review workflows, or audit trails, which means it does not scale to multi-annotator teams without a lot of manual file-shuffling.
According to Mercy: “No collaboration features, and it’s not suitable for large teams or massive datasets.” From a QA standpoint, VIA offers no review workflows, no inter-annotator agreement measurement, and no audit trails. It is a research annotation tool, fast, local, and honest about its scope.
What VIA does have, which often gets overlooked, is institutional credibility. It was built at one of the leading computer vision research groups in the world, is actively maintained on GitHub as of 2026, and has been used well beyond computer vision, in humanities, history, and zoology research. For academic annotation work, that lineage carries weight.
VIA is a genuinely good choice for a solo project or a first pass at a dataset. If your project later needs more annotators, more oversight, or subject-matter expertise VIA doesn’t help with, that’s a different kind of problem than VIA was built to solve. We’ve worked with research teams at that stage, including:
Segments.ai - The Specialist LiDAR Annotation and Sensor Fusion Platform
Segments.ai is not trying to be a general-purpose annotation tool. It is built specifically for teams annotating multi-sensor data: LiDAR point clouds, camera imagery, and the fusion of both simultaneously. For teams working in autonomous driving, robotics, and ADAS development, it handles the annotation challenge that general-purpose tools consistently handle poorly.
Joy, our project coordinator, recommends it specifically when the data has a third dimension: “Good choice when segmentation quality matters, specifically point clouds and multi-sensor fusion. The interface for projecting 3D LiDAR onto 2D images is seamless.”
The standout capability is the simultaneous 2D/3D labeling interface, camera and LiDAR views synchronized so annotations made in 3D project automatically onto the 2D camera image, eliminating what is otherwise a significant and error-prone manual step. Joy notes that “their 2D to 3D label projection is a massive time-saver for robotics.”
From a technical standpoint, Sai, our System Administrator, rates it as “the best mid-tier tool for sensor fusion and LiDAR annotation,” specifically calling out the “camera + LiDAR synchronized view with automatic 2D projection of 3D annotations” as exactly what ADAS and autonomous driving projects require. The Python SDK supports pipeline automation, and the platform ingests standard LiDAR formats including .pcd and .bin files. Export formats cover OpenLABEL (ASAM), KITTI, NuScenes, and custom JSON, the formats that autonomous driving ML pipelines actually use, not generic CV export formats.
Additional capabilities that matter for production annotation programs: label propagation for video sequences, issue tracking for label quality, consensus annotation across up to 30 annotators, and collaborative features that VIA and general open source tools do not offer. The platform also includes batch processing and automated tracking.
The limitations are also clearly stated by the team. Joy notes it is “a more specialised and pricey tool, which isn’t ideal for basic 2D image classification projects or smaller teams.” Sai adds that it is “less flexible for non-sensor fusion tasks,” has “fewer integration connectors compared to Label Studio,” and “custom label schemas less flexible than Label Studio.”
The community is smaller than CVAT or Label Studio. On data privacy: Segments.ai’s primary offering is cloud-based, with self-hosting options available, teams handling sensitive data should verify residency requirements directly before committing.
Segments.ai offers four pricing tiers: a free trial, a free academic license, a Team plan at $9,600 per year, a Scale plan for advanced multi-sensor setups, and an Enterprise plan for custom automation requirements. It is a commercial specialist platform, not open source, included in this review because it represents the category of 3D annotation work that general-purpose tools cannot yet handle at production quality, and because multiple members of the HITL team use it on real client projects.
Segments.ai is the right choice for engineering teams annotating LiDAR point clouds, camera-LiDAR sensor fusion data, or autonomous driving and robotics datasets. As Joy, our QA specialist, puts it: “If your project has a Z-axis, start here.”
The platform handles the tooling – sync, projection, export formats. It doesn’t do the labeling itself. If you’re weighing in-house hires against an outside team for the actual annotation work, our annotation services also cover LiDAR and sensor fusion data, alongside geospatial, medical, industrial, and agricultural work. Book a call to talk through your project.
Quick Comparison: VGG VIA vs Segments.ai for LiDAR annotation and Research Annotation
VGG Image Annotator (VIA) | Segments.ai | |
Best for | Solo research, quick image annotation | LiDAR, sensor fusion, autonomous driving |
Data types | Images, audio, video | LiDAR, camera, sensor fusion, 3D point cloud |
Open source | ✅ BSD-2 license, fully free | ❌ Commercial platform (free trial available) |
Self-hosting | ✅ Runs locally in browser | ⚠ Cloud-primary; self-hosting available |
Collaboration | ❌ Single annotator | ✅ Team collaboration, issue tracking, consensus |
AI-assisted annotation | ❌ None | ⚠ Moderate — some AI-assisted pre-labeling |
Export formats | CSV, JSON | OpenLABEL, KITTI, NuScenes, custom JSON |
Setup | Zero — open HTML file in browser | Cloud-first, straightforward onboarding |
Pricing | Free | Team plan $9,600/year; free trial + free academic license available |
Ideal team size | Solo to small | Small to medium engineering teams |
HITL verdict | Best zero-cost tool for quick research annotation | Best specialist tool for 3D and sensor fusion |
Which LiDAR Annotation Tool or Research Tool Should You Choose?
If you are a solo researcher, an academic team, or an engineer who needs to annotate a dataset quickly without any infrastructure overhead, VGG VIA is the right starting point. It handles the annotation types that most basic computer vision projects require, exports to formats your ML pipeline already reads, and costs nothing. The trade-off is that it does not scale, when your project grows to require team collaboration or QA workflows, you will need a different tool.
If your project involves LiDAR point clouds, camera-LiDAR sensor fusion, or multi-sensor annotation for autonomous driving or robotics applications, Segments.ai is the right choice. It is a paid specialist platform, but it fills a gap that no free tool in this category fills adequately at production quality. The automatic 2D projection of 3D annotations alone saves significant annotation time on sensor fusion projects. For teams where cost is the primary constraint, CVAT offers basic 3D point cloud support at no cost, covered fully in Episode 1 of this series.
If you are working at enterprise scale, across multiple modalities, or need built-in QA workflows and self-hosting for data privacy compliance, neither of these tools is the right answer, that is what the platforms covered in our enterprise annotation review are built for.
FAQ
Is VGG Image Annotator (VIA) free for commercial use? Yes. VIA is released under the BSD-2 clause license, which permits both academic and commercial use at no cost.
Does VIA support LiDAR or point cloud data? No. VIA supports images, audio, and video only, it has no point cloud or 3D annotation capability. For LiDAR work, see Segments.ai or CVAT below.
How much does Segments.ai cost? Segments.ai’s Team plan starts at $9,600/year, with a free trial and a free academic license also available. Scale and Enterprise pricing is custom.
Is Segments.ai still independent? No. Segments.ai was acquired by Uber in October 2025 and now operates under Uber AI Solutions.
