Skip to main content

🎉 We released a Spotlight OSS Version! ⭐ Star it on Github

Version: 1.0.0

Create image embeddings with Towhee

We use the towhee library to create an embedding for a an image dataset.

Use Chrome to run Spotlight in Colab. Due to Colab restrictions (e.g. no websocket support), the performance is limited. Run the notebook locally for the full Spotlight experience.

Open In Colab
  • df['image'] contains the paths to the images in the dataset


Imports and play as copy-n-paste functions

# Install dependencies
# Imports
!pip install renumics-spotlight towhee datasets
#Play as copy-n-paste functions
#@title Play as copy-n-paste functions

import datasets
from towhee import pipeline, DataCollection
from renumics import spotlight
import pandas as pd

def towhee_embedding(df, modelname='towhee/image-embedding-swin-base-patch4-window7-224', image_name='image'):
dc = DataCollection(df[image_name])
embedding_pipeline = pipeline(modelname)
dc_embedding =

df_emb = pd.DataFrame()

return df_emb

Step-by-step example on CIFAR-100

Load CIFAR-100 from Huggingface hub and convert it to Pandas dataframe

dataset = datasets.load_dataset("renumics/cifar100-enriched", split="train")
df = dataset.to_pandas()

Compute embedding with vision transformer from Huggingface

df_emb=towhee_embedding(df, modelname='towhee/image-embedding-swin-base-patch4-window7-224')
df = pd.concat([df, df_emb], axis=1)

Reduce embeddings for faster visualization

import umap
import numpy as np
embeddings = np.stack(df['embedding'].to_numpy())
reducer = umap.UMAP()
reduced_embedding = reducer.fit_transform(embeddings)
df['embedding_reduced'] = np.array(reduced_embedding).tolist()

Perform EDA with Spotlight

df_show = df.drop(columns=['embedding', 'probabilities']), port=port, dtype={"image": spotlight.Image, "embedding_reduced": spotlight.Embedding})