DART: Training-Free Framework Converts SAM to Real-Time Detector
rsasaki0109 · x · 2026-08-29
DART (Detect Anything in Real Time) is a training-free framework designed to convert frontier object detection models (like SAM) into real-time, multi-class, open-vocabulary detectors.
Performance:
- Achieves 15.8 FPS on a single RTX 4080 for 4-class, 1008px resolution images.
- Scores 55.8 AP on the COCO val2017 dataset (80 classes).
The framework addresses the efficiency bottleneck of existing models like SAM in real-time multi-class detection applications.
Related event: DART Turns SAM into a Real-Time Open-Vocabulary Detector Without Training(2 posts)→
More from Multimodal
- The math: 5 rerolls at 480p plus a 4K upscale costs $5 vs $9 at 720p — HeyAmit_ · 2026-10-15
- Dreamina workflow: generate at 480p, upscale to 4K, cut video costs 44% — HeyAmit_ · 2026-10-15
- One-prompt GPT-6 Astra writes two Palestrina-style motets in LilyPond — deanwball · 2026-10-15
- DiffSynth-Studio Adds OPD Workflow: Few-Step Diffusion Training in ~2 GPU-Hours — bdsqlsz · 2026-10-15
- AI film and TV hit the market: Prime Video picks up AI spin-off, pilots screen at Mipcom — lmoroney · 2026-10-15
- Ramayana film taps DNEG's massive VFX and AI-assisted production for global release — CurieuxExplorer · 2026-10-15