WR92 COURSE

No Hugging Face GPU needed · Detection runs in your browser and does not use HF GPU minutes or generation calls. Device, network and model downloads can still limit it.目标检测不消耗 Hugging Face GPU · 检测在本机浏览器完成,没有 ZeroGPU 分钟或生成次数限制,但仍受设备、网络和模型下载影响。 ZeroGPU policy →ZeroGPU 官方规则 →

01 / LEARN · IMAGE UNDERSTANDING学习 · 图像理解

AI-Powered
Object Detection

Pick a COCO example or upload your own image. Then choose a pretrained detector and let AI answer: what objects are here, and where are they?

Task Object detection Data COCO validation examples Models RT-DETR · DETR · YOLOS Tiny
Real COCO example input showing a cat and dog indoors, before running object detection
REAL IMAGE INPUT真实输入图片COCO val2017 · 000000189806 · not model outputCOCO val2017 · 000000189806 · 不是模型预测图
1

Choose an image

Start with a tested COCO example, or try a photo of your own.

Start here · COCO examples

Four images are drawn at random from 100 bundled COCO validation examples.

or try your own

Useful for seeing where the model generalizes well—and where it does not.

2

RT-DETR is the recommended classroom default.

3
0.50

Higher values only keep predictions the model is more confident about.

Choose a COCO example or upload your own image.

AI prediction

What did the detector find?

Waiting

Choose an image first. The original image appears here before detection.

No detections yet.
Object Confidence Bounding box
Run the detector to see details.

What you are doing

Use a learned AI capability instead of rebuilding the traditional task from scratch.

COCO contains everyday-object images with annotations. The selected pretrained detector has already learned the object-detection task from labeled data. In this mini-lab, you can switch models and compare how their predictions differ on the same image.

COCO example or your image → Selected pretrained detector → Boxes + labels + confidence

Compare models: keep the image fixed, switch detector, and observe how predictions, confidence, and speed change.