DART | Mehmet Kerem Turkcan

DART turns a promptable frontier vision model into a real-time multi-class open-vocabulary detector. The project targets a practical deployment gap: Strong promptable segmentation models can describe almost anything, but repeated per-class inference is too slow for many real-world systems.

DART qualitative detections in a crowded street scene

The public release includes code, benchmarks, TensorRT deployment paths, distilled student backbones, and Hugging Face weights.

Role: Creator and maintainer
Adoption: 300+ GitHub stars and 43 forks as of July 2026
Keywords: Open-vocabulary detection, real-time inference, backbone sharing, batched multi-class decoding, TensorRT FP16 optimization, adapter distillation, release engineering

Links: GitHub, arXiv, Hugging Face.

DART is a training-free framework that converts a promptable frontier vision model into a real-time multi-class open-vocabulary detector by sharing class-agnostic visual backbone computation, batching decoding, and optimizing inference for deployment.

References

2026