X-AnyLabeling: A lightweight, efficient, and unified cross-platform desktop application for annotating text, image, video, point cloud and multimodal data, combining versatile built-in tools with state-of-the-art AI models and flexible multi-format export.
Images to inference with no labeling (use foundation models to train supervised models).
A ComfyUI custom node designed for advanced image background removal and object, face, clothes, and fashion segmentation, utilizing multiple models including RMBG-2.0, VRMBG-3.0, FeyNoBg, Finegrain Box Segmenter, INSPYRENET, BEN, BEN2, BiRefNet, Lucida, SDMatte, SAM, SAM2, SAM3, Florence-2, YOLOv8/YOLO11, Big-Lama, and GroundingDINO.
All-in-one training for vision models (YOLO, ViTs, RT-DETR, DINOv3): pretraining, fine-tuning, distillation.
solo-learn: a library of self-supervised methods for visual representation learning powered by Pytorch Lightning
Research and Production Oriented Speaker Verification, Recognition and Diarization Toolkit
Turn your cameras into Home Assistant sensors with a local AI: garage door open/closed, people and cars, water, gas and power meters read by any camera or ESP32-CAM. 100% local, any CPU.
CVE-2026-67276 RouterOS SSH public-key authentication bypass lab PoC
MariaDB 13.0.1-rc RCE lab β priv-esc + heap UAF + JOP chain to system() as uid 999(mysql) on stock Docker image. Found with RAPTOR and raptor-loop-hunt.
Full-circle image roll estimation with calibrated abstention. Open weights, Apache-2.0. By ORTUS AI, the team behind CHEQIT.
β
HFHPE: High-Angle Robust Fast Head Pose Estimation. One-pass yaw/pitch/roll head pose (biternion cos/sin + von Mises ΞΊ) distilled from a DINOv3 teacher into compact 64px students, robust to extreme angles and heavy occlusion.