grounding-dino-base
by IDEA-Research
Open-set detection using natural language descriptions
IDEA-Research/grounding-dino-basemixpeek://image_extractor@v1/idea_grounding_dino_base_v1Overview
Grounding DINO combines a DINO-style detection transformer with grounded language understanding for open-set object detection. It achieves 52.5 AP on COCO with zero training data on COCO, and 56.7 AP when fine-tuned.
On Mixpeek, Grounding DINO enables detecting any object by describing it in text. Combined with segmentation models like SAM, it provides a powerful detect-then-segment pipeline.
Architecture
DINO-style detection transformer with Swin backbone, enhanced with text-grounding modules for open-vocabulary detection. Swin-B variant achieves 56.7 AP on COCO.
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/image.jpg" },feature_extractors: [{name: "object_detection",version: "v1",params: { model_id: "IDEA-Research/grounding-dino-base" }}]});
Capabilities
- Zero-shot detection: 52.5 AP on COCO without COCO training data
- Natural language object descriptions as prompts
- Fine-tuned detection: 56.7 AP (Swin-B)
- Pairs with SAM for detect-then-segment pipelines
Use Cases on Mixpeek
Benchmarks
| Dataset | Metric | Score | Source |
|---|---|---|---|
| COCO val2017 (zero-shot) | AP | 48.4 | Liu et al., 2024 — Table 1 |
| RefCOCO (val) | Accuracy | 89.2% | Liu et al., 2024 — Table 3 |
Performance
Common Pipeline Companions
Specification
Research Paper
Grounding DINO: Marrying DINO with Grounded Pre-Training for Open-Set Object Detection
arxiv.orgBuild a pipeline with grounding-dino-base
Add this model to a processing pipeline alongside other extractors. Combine with retrieval stages for end-to-end search.
Open Pipeline Builder