Dataset Card for Emojis

This is a FiftyOne dataset with 1816 samples.

Installation

If you haven't already, install FiftyOne:

pip install -U fiftyone

Usage

import fiftyone as fo
import fiftyone.utils.huggingface as fouh

# Load the dataset
# Note: other available arguments include 'max_samples', etc
dataset = fouh.load_from_hub("jamarks/emojis")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

• Curated by: Jacob Marks
• Language(s) (NLP): en
• License: cc-by-4.0

Dataset Sources

• Demo: https://try.fiftyone.ai/datasets/emojis/samples

Dataset Creation

Curation Rationale

Emojis sit at the intersection between textual and visual, providing a fascinating test-bed for exploring multimodal search and reranking techniques. This dataset was constructed to facilitate these experiments. For connected projects, check out:

• Emoji Search CLI Library
• Semantic Emoji Search Plugin for FiftyOne

Source Data

Samples in this dataset were constructed from rows in the Kaggle Full Emoji Image Dataset

Data Collection and Processing

The base64-encoded images in the original csv were upscaled by 10x using Real-ESRGAN.

OpenAI's CLIP-VIT-B/32 model was used to embed these images (vision encoder), the emoji names (text encoder), and the unicode sequences (text encoder). These embeddings were used to construct Brain Runs for performing similarity and semantic searches, as well as visualizing the structure of the dataset using UMAP dimensionality reduction.

Dataset Card Authors

Jacob Marks