Spaces:
Runtime error
Runtime error
File size: 3,252 Bytes
a3ee979 2eca6de a3ee979 54a3362 a3ee979 a0fbf80 a3ee979 26b1205 54a3362 a3ee979 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 |
from typing import Optional
import gradio as gr
import torch
from src.nn import CaSED
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%3A2306.00917-B31B1B.svg"/>
</a>
<a href="https://altndrr.github.io/vic/" 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 = CaSED().to(DEVICE).eval()
def vic(filename: str, alpha: Optional[float] = None):
# get the outputs of the model
vocabulary, scores = model(filename, alpha=alpha)
confidences = dict(zip(vocabulary, scores))
return confidences
def resize_image(image, max_size: int = 256):
"""Resize image to max_size keeping the aspect ratio."""
width, height = image.size
if width > height:
ratio = width / height
new_width = max_size * ratio
new_height = max_size
else:
ratio = height / width
new_width = max_size
new_height = max_size * ratio
return image.resize((int(new_width), int(new_height)))
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="./artifacts/examples/",
allow_flagging='never',
theme=gr.themes.Soft(),
thumbnail="https://altndrr.github.io/vic/assets/images/method.png",
)
demo.launch(share=False)
|