OpenCV implemented DEIMv2 with two backbone options, HGNetV2 and DINOv3, for the head detector.
Public source
Publisher name
Public post
🎯 Robust Face Alignment at Extreme Head Angles I demoed HRFFA, a high-angle robust, fast face alignment model, and it is genuinely impressive at angles that usually bre…
Company
OpenCV
OpenCV is the largest computer vision library in the world.
- Industry
- Non-profit Organizations
- Company size
- 51–200 employees
About OpenCV
OpenCV is the world’s leading open-source computer vision library and one of the largest open-source computer vision frameworks, with more than 40 million monthly downloads. OpenCV provides developers, researchers, and organizations with tools for real-time image processing, computer vision, and AI-powered perception across robotics, industrial automation, healthcare, manufacturing, security, consumer devices, education, and research.
See moreLatest activity
Latest activity from OpenCV
15 signals
Products & Services
OpenCV demoed HRFFA, a high-angle robust face alignment model running at about 8.7 fps with OpenCV's DNN and ONNX Runtime backend.
Products & Services
OpenCV implemented a MIT-licensed face alignment model with results at extreme head angles.
Products & Services
OpenCV implemented a 256-resolution aligner with a CNN-based aligner built on HGNetV2 and a ViT aligner.
Discover more
Similar signals
Similar public activity from other companies.
Products & Services
Roboflow
Roboflow released RFDETR 1.10.0, a training-focused Open Source release with a 10k-image dataset trained for 10 epochs on an L4 dataset going from 55 minutes to 31 minutes.
Products & Services
Lightly
Lightly released the new LT-DETRv2 model line-up for object detection and instance segmentation tasks with LightlyTrain, an open-source framework for self-supervised learning and downstream fine-tuning.
Products & Services
Hugging Face
Hugging Face gathered 41 OCR models into four collections for Japanese manga, Swedish handwriting, and Arabic print.
Products & Services
Deep Learning for Earth Observation (DL4EO)
Deep Learning for Earth Observation (DL4EO) released Oriented-Det v0.2.0, a one-stage detector with a 3× decoded-rIoU checkpoint reaching 81.58% eval-val mAP50 on DOTA.
Products & Services
Mozilla Data Collective