YOLO-World-L
by AILab-CVC
Real-time open-vocabulary object detection with text prompts
AILab-CVC/YOLO-World-Lmixpeek://image_extractor@v1/tencent_yoloworld_large_v1Overview
YOLO-World extends the YOLO detector family with open-vocabulary detection via vision-language modeling. Users specify objects to detect with text prompts; the model finds them zero-shot at real-time speeds (52 FPS on V100).
On Mixpeek, YOLO-World enables detecting arbitrary objects in video and images using natural language, without retraining for each new category.
Architecture
YOLO backbone with Re-parameterizable Vision-Language Path Aggregation Network (RepVL-PAN). Uses region-text contrastive loss and a prompt-then-detect paradigm where vocabulary is embedded as model parameters for fast inference.
Mixpeek SDK Integration
import { Mixpeek } from "mixpeek";const mx = new Mixpeek({ apiKey: "API_KEY" });await mx.collections.ingest({collection_id: "my-collection",source: { url: "https://example.com/video.mp4" },feature_extractors: [{name: "object_detection",version: "v1",params: { model_id: "AILab-CVC/YOLO-World-L" }}]});
Capabilities
- Open-vocabulary detection with text prompts
- 52 FPS on V100 (real-time)
- 35.4 AP on LVIS zero-shot
- Supports image-prompted detection
- ONNX and TFLite INT8 export
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| LVIS (zero-shot) | AP | 35.4 | Cheng et al., 2024 — Table 1 |
| COCO val2017 | AP | 45.7 | Cheng et al., 2024 — Table 2 |
Performance
Specification
Research Paper
YOLO-World: Real-Time Open-Vocabulary Object Detection
arxiv.orgBuild a pipeline with YOLO-World-L
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Open Pipeline Builder