tensorrt
PulseAugur coverage of tensorrt — every cluster mentioning tensorrt across labs, papers, and developer communities, ranked by signal.
9 day(s) with sentiment data
-
Ultralytics YOLO Evolution: From YOLOv5 to YOLO27 Detailed in New Paper
A comprehensive paper reviews the evolution of Ultralytics' YOLO object detection models, detailing advancements from YOLOv5 through the latest YOLO27. YOLO27 introduces a dual-architecture strategy, with compact versio…
-
VPEngine framework boosts robotic vision inference speed by 3x
Researchers have developed VPEngine, a novel framework designed to optimize GPU usage for robotic vision tasks. This system utilizes a shared foundation model to extract image representations, which are then efficiently…
-
TokenMask simplifies vision transformer segmentation, boosting efficiency
Researchers have developed TokenMask, a novel method for vision transformer segmentation that operates directly in the token space, eliminating the need for dense spatial feature map reconstruction. This approach simpli…
-
AI framework enables real-time fruit detection on edge hardware
Researchers have developed a framework for real-time fruit detection and video analytics on embedded hardware, specifically the NVIDIA Jetson Orin Nano Super. The system utilizes a lightweight YOLO26s detector trained o…
-
New 'woma' foundation model sets real-time endoscopy standard
Researchers have developed "woma," a real-time foundation model for gastrointestinal endoscopy, trained without labels on approximately one million endoscopy frames. This model can be fine-tuned for specific tasks, such…
-
Running multiple AI models on one GPU: challenges and considerations
This article explores the challenges of running multiple AI models on a single GPU, focusing on potential conflicts and performance degradation. It highlights key questions to consider regarding model interference, syst…
-
LLM inference servers share GPU memory to serve hundreds of users
Serving hundreds of users simultaneously with a single GPU for large language models is achieved by loading the model weights into GPU memory once and sharing them across all requests. The inference server manages this …
-
Task Compiler concept shifts AI development from models to intent
A new concept called the "Task Compiler" proposes a shift from model-centric to task-centric AI development. This approach would allow developers to define desired outcomes and constraints, such as accuracy and latency,…
-
MiniMax-H3 VAE optimized for ComfyUI boosts speed up to 1.7x
A new implementation of the MiniMax-H3 variational auto-encoder (VAE) is now available for ComfyUI, optimized with TensorRT. This version aims to significantly improve processing speed, with reported gains of up to 1.7 …
-
Serving YOLOv8 with NVIDIA Triton via ONNX and TensorRT
This article details how to serve the YOLOv8 object detection model using NVIDIA Triton Inference Server. It explains the process of converting the ONNX format of YOLOv8 to TensorRT, a high-performance inference optimiz…
-
Open-source SeedVR2+TensorRT Studio offers faster local video upscaling
A new open-source application called SeedVR2+TensorRT Studio has been released, offering local, GPU-accelerated video restoration and upscaling. This tool utilizes SeedVR2 and TensorRT for faster processing, particularl…
-
AWS, NVIDIA, and Heidi Health cut ASR inference costs by 75%
AWS, NVIDIA, and Heidi Health collaborated to reduce automatic speech recognition (ASR) inference costs by 75% on Amazon EC2 instances. By implementing NVIDIA MPS with the NVIDIA Triton Inference Server, they achieved a…
-
AI frameworks tackle plant disease diagnosis and fruit classification
Researchers have developed advanced AI frameworks for agricultural applications, focusing on plant disease diagnosis and fruit classification. The first study introduces H²MAF, which fuses vision models like EfficientNe…
-
Low-cost AI system detects fabric defects on edge devices
Researchers have developed a two-stage fabric defect detection system designed for low-cost deployment on edge devices like the NVIDIA Jetson Nano. The system uses a lightweight autoencoder for initial screening, follow…
-
AI pipeline enhances child labor detection with improved accuracy
Researchers have developed a real-time edge vision pipeline designed to assist in combating child labor by detecting children and estimating their ages. This system, built as a research prototype, significantly improves…
-
Nvidia Jetson DLA cores enable near-zero overhead classification in real-time pipelines
Researchers have developed a novel five-step methodology to enable the deployment of classification models on NVIDIA Jetson DLA cores, overcoming challenges with strict operator constraints and quantization incompatibil…
-
Ollama v0.32.6 boosts Qwen 3.5 speed on Apple Silicon, improves OpenAI compatibility · 4 sources tracked
Ollama has released version 0.32.6, significantly improving the performance of the Qwen 3.5 model on Apple Silicon Macs through the MLX engine and speculative decoding. This update also enhances compatibility with OpenA…
-
PixOOD pipeline optimized for real-time anomaly segmentation in autonomous vehicles
Researchers have developed an efficient anomaly segmentation pipeline called PixOOD, designed for real-time deployment on embedded hardware in autonomous vehicles and railway systems. This new approach optimizes the Ney…
-
Lightweight AI model identifies raptor species for wind turbine safety
Researchers have developed a lightweight image classification system for identifying raptor species on edge devices, specifically for wind turbine collision mitigation. The system utilizes knowledge distillation to trai…
-
NVIDIA TensorRT adds build monitoring and cancellation features
NVIDIA has introduced new capabilities for its TensorRT engine, allowing users to monitor and cancel long-running engine builds directly through Python or C++ interfaces. This enhancement provides greater control over l…