AI Briefing
KO

BloodshotNet Released

·2026.04.25 00:00

Key point

BloodshotNet has released an open-source model for blood detection along with its dataset, weights, and CLI.

Details

BloodshotNet has released an open-source model for blood detection.

The components released alongside it are as follows.

  • dataset: 23,000+ annotated images from forensic scenes, UFC footage, horror/gore films, and surgical content
  • model weights: YOLO26 small and nano variants, licensed under AGPL-3.0
  • CLI: a command-line tool that analyzes images, folders, and videos in one go

As published metrics, they presented precision of about 0.8, recall of about 0.6, and 40+ FPS on CPU for the small model.

The team explained that for video-level filtering, a sliding window over 5-10 second clips is more effective than perfect per-frame judgments. They also noted that text-prompt-based open-vocabulary models struggled to capture blood's irregular patterns, and that the YOLO26 + ProgLoss + STAL combination performed better, especially on small bloodstains and droplets.

computer-visionopen-sourceobject-detectioncontent-moderation
Read originalShare on Threads

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.

Related

RedditOpen-Weight CLM Released as Jev Alternative, Agent Decision-Making 4x FasterPyTorchKRTrillion Labs Releases GravityOCR, an OCR Model Applying Self-Speculative DecodingHacker NewsJevper Released: A Jev Classification Wrapper for OpenAI-Compatible Models with Automatic Logprobs DetectionGeekNewsRootless Linux Sandbox 'Drop' Released for Isolating AI Agents and Third-Party ProgramsPyTorchKRPostgreSQL Extension pgjev Released, Supporting Natural Language Filtering Without EmbeddingsTLDR AIJared Palmer Releases 'Kev', a Lightweight Decision-Making Model Based on Qwen3.5