图像分类
使用设备端视觉模型对图像进行分类（前 5 个标签及分数）。图像不会离开您的设备；模型在首次使用时下载。
First use downloads the ViT image-classification model (Xenova/vit-base-patch16-224, quantized build) from Hugging Face. After the download, classification runs entirely on your device and your image is never uploaded anywhere.
使用方法
- Choose an image file (on a phone you can also take a photo).
- Press Classify. The first run downloads the model.
- The top 5 predictions appear with their scores.
注意事项
The model is Xenova/vit-base-patch16-224 (the 1,000 ImageNet-1k classes). Only the first use downloads the model files from Hugging Face; classification itself runs on-device with Transformers.js and your image is never uploaded. Labels are English class names, and scores are the model's confidence, not a guarantee of correctness.
常见问题
- Why are some labels odd or wrong?
- ImageNet classes are very fine-grained (dog breeds, car models, …) and a general model misses on subjects outside its strengths. Treat results as indicative.
- The download is slow or fails
- The model download can take a while. Retry on a good connection. Downloaded files are cached by the browser, so later visits start much faster.