open-vocabulary object detection
YOLO-World is a real-time open-vocabulary object detector that finds arbitrary categories from text prompts without retraining, built on a YOLO-style architecture.
Solves the same problem with a different architecture or at a different layer. Expect to rewrite the integration.
Pick it if you are already on PaddlePaddle and want a full detection/segmentation/tracking toolkit, accepting fixed classes instead of text prompts.
Object Detection toolkit based on PaddlePaddle. It supports object detection, instance segmentation, multiple object tracking and real-time multi-person keypoint detection.
Choose it for the most maintained, export-friendly real-time detector stack when fixed-class training is acceptable.
Ultralytics YOLO26, YOLO11, YOLOv8 — object detection, instance segmentation, semantic segmentation, image classification, pose estimation, object tracking
Take it for Apache-2.0 real-time DETR backbones with segmentation when closed-set fine-tuning is enough.
RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]
Choose it to stay inside the PaddlePaddle ecosystem with YOLO-style closed-set detectors.
🚀🚀🚀 YOLO series of PaddlePaddle implementation, PP-YOLOE+, RT-DETR, YOLOv5, YOLOv6, YOLOv7, YOLOv8, YOLOv10, YOLO11, YOLOX, YOLOv5u, YOLOv7u, YOLOv6Lite, RTMDet and so on. 🚀🚀🚀
Choose it for an actively maintained NMS-free transformer detector under Apache-2.0, but with closed vocabularies only.
[CVPR 2024] Official RT-DETR (RTDETR paddle pytorch), Real-Time DEtection TRansformer, DETRs Beat YOLOs on Real-time Object Detection. 🔥 🔥 🔥
Pick it for the latest DETR regression refinements in real-time closed-set detection.
D-FINE: Redefine Regression Task of DETRs as Fine-grained Distribution Refinement [ICLR 2025 Spotlight]
Use it as a DETR research platform to train custom transformer detectors, accepting a less turnkey pipeline.
detrex is a research platform for DETR-based object detection, segmentation, pose estimation and other visual recognition tasks.
Choose it for open-vocabulary detection with language-aware fusion in a DETR-style framework, though it is less real-time oriented and lightly maintained.
OV-DINO: Unified Open-Vocabulary Detection with Language-Aware Selective Fusion
Pick it for a battle-tested closed-set detector with the widest export and deployment tooling.
Ultralytics YOLOv5 in PyTorch for object detection, instance segmentation, classification, training, and export.
Choose it for NMS-free real-time latency on fixed classes, without open-vocabulary prompting.
YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]
Pick it for faster DETR convergence when training closed-set detectors.
[CVPR 2025] DEIM: DETR with Improved Matching for Fast Convergence
An earlier or more limited way to do the job. Still the right call when you need something small, proven, or CPU-only.
Use it for a config-driven closed-set detection zoo, but it has gone unmaintained for two years and has no open-vocabulary weights.
OpenMMLab Detection Toolbox and Benchmark
Only for reproducing end-to-end DETR research; archived, slow to converge, and fixed-class.
End-to-End Object Detection with Transformers
Only for legacy center/keypoint-based detection code; unmaintained for over three years and class-limited.
Object detection, 3D detection, and pose estimation using center point detection:
For reproducing the YOLOv7 recipe; unmaintained for two years and requires fixed classes.
Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
For industrial fixed-class deployment on a 2022-era model that is no longer updated.
YOLOv6: a single-stage object detection framework dedicated to industrial applications.
Only for legacy YOLOv3 pipelines and tiny-model exports; much lower accuracy and no text prompts.
PyTorch implementation of YOLOv3, YOLOv3-SPP, and YOLOv3-tiny for real-time object detection with training, validation, inference, and multi-format export.
Consider it for joint open-vocabulary segmentation and detection, but it is unmaintained and not real-time.
[ICCV 2023] Official implementation of the paper "A Simple Framework for Open-Vocabulary Segmentation and Detection"
These projects were analysed and named YOLO-World among their alternatives. The relationship is not symmetric — how YOLO-World rates them is a separate judgement, made when YOLO-World is analysed in its own right.
calls YOLO-World “Choose when you need real-time open-vocabulary detection from arbitrary text prompts, not YOLOv7's fixed COCO-style classes.”
Implementation of paper - YOLOv7: Trainable bag-of-freebies sets new state-of-the-art for real-time object detectors
calls YOLO-World “Choose it if you need open-vocabulary detection with text prompts beyond fixed classes.”
YOLOv10: Real-Time End-to-End Object Detection [NeurIPS 2024]
calls YOLO-World “Choose when you need open-vocabulary object detection within a YOLO-like framework.”
PyTorch implementation of YOLOv3, YOLOv3-SPP, and YOLOv3-tiny for real-time object detection with training, validation, inference, and multi-format export.
calls YOLO-World “choose it when you need open-vocabulary detection with arbitrary text-defined classes rather than fixed-class fine-tuning.”
RF-DETR is a real-time object detection and segmentation model architecture developed by Roboflow, SOTA on COCO, designed for fine-tuning. [ICLR 2026]
calls YOLO-World “Choose when you need open-vocabulary detection driven by text prompts; CenterNet only predicts fixed training classes.”
Object detection, 3D detection, and pose estimation using center point detection: