#Ultralytics
Ultralytics YOLO27, YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking

🔗 https://github.com/ultralytics/ultralytics
November 16, 2026 at 11:15 PM
Ultralytics 8.4.212 adds OCR for Korean, Cyrillic, Arabic, Devanagari, Thai, Greek, Tamil, and Telugu scripts.
November 15, 2026 at 10:15 PM
Ultralytics YOLO v8.4.209: Concurrent YOLO predictions now apply per-request settings, with fixes for validation, exports, and device-specific inference.
November 15, 2026 at 8:45 AM
No benchmarks, licence 'other', empty card — reads as a stock Ultralytics repackage. The x-plus-P6 bet for high-res pose is fair, but hard to see an Ascend team choosing this over upstream until CANN conversion is documented; Modelers hosting alone shouldn't earn the download.
YOLOv8x-Pose-P6 Lands on Modelers.cn: Bigger Keypoints, Sparse Model Card
Modelers.cn has added museteam/yolov8x-pose-p6, an Ultralytics-style YOLOv8 pose-estimation checkpoint aimed at detecting people and predicting 2D body keypoints in images or video. For developers,…
aichina.news
November 18, 2026 at 7:23 PM
Licence 'other', no provenance — a mirror, not an Ascend port. Still 68.9M params of single-pass 17-keypoint pose, with ONNX export as a credible route through CANN's ATC. The empty card is a nuisance; anyone who has pushed YOLO through ATC knows conversion is the real gate.
YOLOv8x-pose Lands on Modelers.cn: Ultralytics' Heaviest Pose Model Comes to the Ascend Ecosystem
If you build anything that needs to know where a person's limbs are — fitness form checks, sports analytics, gesture control, motion capture — Ultralytics' YOLOv8x-pose is probably already on your…
aichina.news
November 18, 2026 at 7:31 PM
The source, dated, on The Deploy Log:
https://deployedbyai.com/log/f5fa2ffe9863/ultralytics-8-4-212/?k=br

Follow for one line every time AI actually ships, as it ships, 24/7.
November 15, 2026 at 10:15 PM
Empty model card, licence listed as 'other' where Ultralytics ships AGPL-3.0, a creation date in 2026. Still, for Ascend developers a hosted nano pose checkpoint spares a conversion job. Hard to see teams picking this over upstream weights until the licence is settled.
Compact Edge AI: YOLOv8n-Pose Arrives on Modelers.cn
Developers hunting for lightweight computer vision solutions have a new contender on the block. Modelers.cn, the platform dedicated to the Huawei Ascend ecosystem, has recently listed…
aichina.news
November 18, 2026 at 8:48 PM
The source, dated, on The Deploy Log:
https://deployedbyai.com/log/5fe708d97155/ultralytics-yolo-v8-4-209/?k=br

Follow for one line every time AI actually ships, as it ships, 24/7.
November 15, 2026 at 8:45 AM
Empty card, licence 'other' (upstream is AGPL), uploader isn't Ultralytics. The 45.2 mask mAP is real pedigree, but with no .om files or Ascend numbers this is a mirror, not a port. ATC conversion is an afternoon's work — someone will prove me wrong.
YOLOv8x-seg: High-Precision Instance Segmentation Hits the Ascend Ecosystem
For developers embedded in Huawei's Ascend AI ecosystem, Modelers.cn remains a vital resource for discovering tools that bridge the gap between international models and domestic hardware. The…
aichina.news
November 18, 2026 at 6:58 PM
An empty card, AGPL-3.0 tagged 'other', no .om files or CANN notes — this reads as a mirror, not a port. The 3.4M-param nano size is genuinely handy for NPU vision, but hard to see an Ascend deployer choosing these weights over Ultralytics until someone documents the conversion.
Edge AI on a Shoestring: Ultra-Lightweight YOLOv8 Segmentation Arrives on Huawei's Ascend Platform
We are seeing a surge of activity in the Huawei Ascend ecosystem, and developers looking for efficient computer vision solutions have a new entry to consider. Modelers.cn has released the YOLOv8n-seg…
aichina.news
November 18, 2026 at 8:41 PM
Real-time object detection on the Raspberry Pi 5 CPU, no accelerator required. 👀👀

Our latest open-source runtime runs Ultralytics YOLO26n at around 27 inference FPS, helping developers get more performance from existing hardware. learn.arm.com/learning-pat...
November 12, 2026 at 11:25 PM
Och dit noch ...

"Beim #Slopsquatting veröffentlicht ein Angreifer ein schädliches Paket unter einem Namen, den ein KI-Programmierassistent erfunden hat, und verlässt sich darauf, dass Entwickler oder automatisierte Tools es installieren."
www.ultralytics.com/de/glossary/...
Slopsquatting: KI-Angriff auf Abhängigkeiten | Ultralytics
Slopsquatting ist ein Angriff auf die Softwarelieferkette, bei dem von einer KI erfundene Paketnamen ausgenutzt werden, um Entwickler dazu zu verleiten, schädliche Abhängigkeiten zu installieren.
www.ultralytics.com
November 11, 2026 at 1:27 PM
Ultralytics 8.4.208: prediction and tracking default to 0.5 NMS IoU, validation keeps 0.7
November 8, 2026 at 3:15 PM
Ultralytics v8.4.203 adds NXP i.MX95 Neutron NPU export for YOLO models
November 7, 2026 at 2:30 PM
The source, dated, on The Deploy Log:
https://deployedbyai.com/log/02b0bd4a7d2a/ultralytics-8-4-208/?k=br

Follow for one line every time AI actually ships, as it ships, 24/7.
November 8, 2026 at 3:15 PM
ultralytics/yolov5 - Ultralytics YOLOv5 is a state-of-the-art, open-source computer vision project built on the PyTorch framework that has fundamentally transformed how developers and researchers approach real-time object... #OpenSource #Python 🔗 https://ift.tt/GBxoWY0
November 6, 2026 at 12:00 AM
The source, dated, on The Deploy Log:
https://deployedbyai.com/log/1ff43b680f0d/ultralytics-v8-4-203/?k=br

Follow for one line every time AI actually ships, as it ships, 24/7.
November 7, 2026 at 2:30 PM
No mAP, no target part, no variant — an empty card makes this a mirror, not a port. Still, MIT-licensed YOLOv5 on a domestic route is how edge ecosystems standardise. Hard to see this beating a straight Ultralytics pull until Axera documents the quantisation path.
YOLOv5 Shows Up on Modelers.cn — Axera's Edge NPU Take on a Detector Everyone Already Trusts
If you have shipped a real-time vision product in the last five years, you have almost certainly shipped YOLOv5. The single-stage detector from Ultralytics became the default answer to "can we get…
aichina.news
November 5, 2026 at 6:57 PM
Ultralytics 8.4.199 sends raw run name to Ultralytics Platform
November 3, 2026 at 4:15 AM
An MIT tag on weights whose upstream is AGPL-3.0, no mAP, no variants, no provenance. The pull for Axera NPU teams is real — a pre-ported YOLO11 skips the quantisation grind. Hard to see this as a production dependency until the licence and benchmarks are settled.
Axera's YOLO11 Port on Modelers.cn Promises a Shortcut to Edge Vision — Documentation Not Included
Modelers.cn is Huawei's Ascend-oriented model hub, so it is always worth watching what appears there. The latest curio for edge-vision developers is AXERA-TECH/YOLO11, a port of Ultralytics' YOLO11…
aichina.news
November 5, 2026 at 7:17 PM
A same-day stub: no mAP, no latency, no input specs, and an MIT claim at odds with upstream AGPL-3.0. If the licence holds, that is a real win for commercial edge work. Hard to see embedded teams picking it over the Ultralytics weights until numbers land on Axera silicon.
Axera Brings YOLOv8-Pose to the Edge: A Ready-to-Deploy Pose Estimation Model
If you are developing on the edge, pose estimation is a frequent requirement—whether it be for robotic navigation, gesture-controlled interfaces, or fitness tracking. Today, we’re spotlighting a new…
aichina.news
November 5, 2026 at 6:04 PM
The source, dated, on The Deploy Log:
https://deployedbyai.com/log/9644fbe51544/ultralytics-8-4-199/?k=br

Follow for one line every time AI actually ships, as it ships, 24/7.
November 3, 2026 at 4:15 AM
Ultralytics 8.4.193: up to 2x faster CPU ONNX Runtime and OpenVINO prediction
October 31, 2026 at 5:45 AM
Ultralytics 8.4.193: up to 2x faster CPU ONNX Runtime and OpenVINO prediction
October 31, 2026 at 5:45 AM
#後で読む 用メモです→
FastAPIでYOLO(ultralytics)を実行するAPIの構築
FastAPIでYOLO(ultralytics)を実行するAPIの構築 - Qiita
環境構築 Dockerfile FROM nvidia/cuda:12.4.0-devel-ubuntu22.04 ... RUN apt-get update \ && apt-get install -y --no-install-recommends ...
qiita.com
October 31, 2026 at 3:49 PM