Xtreme1

What Xtreme1 is

Xtreme1 is an open-source platform for preparing multimodal training data, with tools to annotate and curate datasets and manage the ontology used for model training. It works with images, 3D LiDAR, and 2D/3D sensor-fusion data. Annotation capabilities include bounding boxes, polygons, polylines, classification, 2D/3D object detection, 3D segmentation, and LiDAR-camera fusion. Data quality monitoring and model-result visualization are also included. Built-in pre-labeling and interactive models assist with detection, segmentation, and classification. The configurable Ontology Center lets teams arrange classes into hierarchies and add attributes. Image annotation links to YOLOR and RITM; fusion workflows link to OpenPCDet and AB3DMOT. RLHF annotation for large language models is a beta feature. The project is Apache 2.0 licensed and identified as an LF AI & Data Foundation project. It can be deployed with Docker Compose on Linux, Windows, or MacOS, or used through web and self-hosted platforms. It exports datasets in COCO, JSON, and Xtreme1 annotation formats. BasicAI is identified in the documentation as the cloud version; its private-cloud deployment starts from $6,600/y, with seats, storage, and model calls customizable.

Who it suits

Xtreme1 suits teams preparing image, LiDAR, or sensor-fusion datasets for model training, especially those needing configurable ontologies and model-assisted labeling. It can also suit organizations seeking open-source software they can deploy themselves.

Limits

The built-in model containers require a Linux server with NVIDIA CUDA and a T4 or similar GPU. BasicAI private-cloud deployment starts from $6,600/y, with seats, storage, and model calls customizable.

Getting started

1. Open the Xtreme1 GitHub project.
2. Check that your system has an AMD64 or ARM64 CPU, at least 2GB RAM, and 10GB or more of free disk space.
3. Install the project with Docker Compose on Linux, Windows, or MacOS.
4. Use the web or self-hosted platform to work with annotation and ontology tools.
5. For built-in model containers, use a Linux server with NVIDIA CUDA and a T4 or similar GPU.
6. For cloud use, see BasicAI; private-cloud deployment starts from $6,600/y.

Verdict

Choose Xtreme1 for multimodal dataset annotation when open-source licensing, ontology management, and self-hosted deployment matter. Consider the BasicAI private-cloud option if you want that deployment model, but its listed starting price is $6,600/y.