Views
No views yet
kittn/eupe_vits16 for transformers.js image feature extraction.config.jsonpreprocessor_config.jsononnx/model.onnxpixel_values shaped [batch, channels, height, width] and returns dynamic last_hidden_state shaped [batch, sequence_length, hidden].transformers.js usage:1import { pipeline, RawImage } from "@huggingface/transformers";
2
3const extractor = await pipeline("image-feature-extraction", "kittn/eupe_vits16-onnx", {
4 device,
5 dtype,
6});
7
8extractor.processor.image_processor.do_resize = false;
9
10const imageData = await RawImage.fromCanvas(imageCanvas);
11const features = await extractor(imageData, { pooling: "none" });
12
13const numRegisterTokens = extractor.model.config.num_register_tokens ?? 0;
14const patchFeatures = features.slice(null, [1 + numRegisterTokens, null]);
15const normalizedFeatures = patchFeatures.normalize(2, -1);