app.py

from typing import Optional

import gradio as gr
import torch
from PIL import Image
from transformers import AutoModel, CLIPProcessor

PAPER_TITLE = "Vocabulary-free Image Classification"
PAPER_DESCRIPTION = """
<div style="display: flex; align-items: center; justify-content: center; margin-bottom: 1rem;">
    <a href="https://github.com/altndrr/vic" style="margin-right: 0.5rem;">
        <img src="https://img.shields.io/badge/code-github.altndrr%2Fvic-blue.svg"/>
    </a>
    <a href="https://altndrr-vic.hf.space" style="margin-right: 0.5rem;">
        <img src="https://img.shields.io/badge/demo-hf.altndrr%2Fvic-yellow.svg"/>
    </a>
    <a href="https://arxiv.org/abs/2306.00917" style="margin-right: 0.5rem;">
        <img src="https://img.shields.io/badge/paper-arXiv.2306.00917-B31B1B.svg"/>
    </a>
    <a href="https://alessandroconti.me/papers/2306.00917" style="margin-right: 0.5rem;">
        <img src="https://img.shields.io/badge/website-gh--pages.altndrr%2Fvic-success.svg"/>
    </a>
</div>


Vocabulary-free Image Classification aims to assign a class to an image *without* prior knowledge
on the list of class names, thus operating on the semantic class space that contains all the
possible concepts. Our proposed method CaSED finds the best matching category within the
unconstrained semantic space by multimodal data from large vision-language databases.

To assign a label to an image, we:
1. extract the image features using a pre-trained Vision-Language Model (VLM);
2. retrieve the semantically most similar captions from a textual database;
3. extract from the captions a set of candidate categories by applying text parsing and filtering;
4. score the candidates using the multimodal aligned representation of the pre-trained VLM to
    obtain the best-matching category.
"""
PAPER_URL = "https://arxiv.org/abs/2306.00917"


device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model = AutoModel.from_pretrained("altndrr/cased", trust_remote_code=True).to(device)
processor = CLIPProcessor.from_pretrained("openai/clip-vit-large-patch14")


def vic(filename: str, alpha: Optional[float] = None):
    images = processor(images=[Image.open(filename)], return_tensors="pt", padding=True)
    outputs = model(images, alpha=alpha)
    vocabulary = outputs["vocabularies"][0]
    scores = outputs["scores"][0].tolist()
    confidences = dict(zip(vocabulary, scores))

    return confidences


demo = gr.Interface(
    fn=vic,
    inputs=[
        gr.Image(type="filepath", label="input"),
        gr.Slider(
            0.0,
            1.0,
            value=0.5,
            label="alpha",
            info="trade-off between the text (left) and image (right) modality",
        ),
    ],
    outputs=[gr.Label(num_top_classes=5, label="output")],
    title=PAPER_TITLE,
    description=PAPER_DESCRIPTION,
    article=f"Check out <a href={PAPER_URL}>the original paper</a> for more information.",
    examples="./examples/",
    allow_flagging="never",
    theme=gr.themes.Soft(),
    thumbnail="https://altndrr.github.io/vic/assets/images/method.png",
)

demo.launch(share=False)