image
imagewidth (px)
18
939
item_ID
stringlengths
13
13
query
stringclasses
1 value
title
stringlengths
1
1.78k
position
int64
0
0
id-0002216949
None
Ransom
0
id-0006128513
None
Caravan to Vaccares
0
id-0006338658
None
Chronicles of Wasted Time: The Infernal Grove v. 2
0
id-0006483682
None
Dragonshadow
0
id-0006511252
None
Flashman (The Flashman Papers, Book 1) (The Flashman Papers)
0
id-0006530451
None
The Black Box
0
id-0007138024
None
The Megalithic European: The 21st Century Traveller in Prehistoric Europe
0
id-0007145578
None
Naked Empire - The Sword Of Truth, Book 8
0
id-0007156073
None
30-Minute-a-Day Body Challenge
0
id-0007172354
None
Q Pootle 5
0
id-0007177429
None
People of the Book
0
id-000719479X
None
The Long March
0
id-0007214324
None
The Broken Souls
0
id-0007262620
None
For the Love of Julie: A nightmare come true. A mother's courage. A desperate fight for justice.
0
id-0007263783
None
Please Don't Make Me Go: The True Story of the Little Boy Who Couldn't Be Beaten
0
id-0007289944
None
Tattoo Sourcebook
0
id-000729204X
None
ELSPETH HUXLEY: A Biography
0
id-0007457138
None
BIG NATE BOREDOM BUSTER 2 PB
0
id-0007583540
None
Collins Very First French Dictionary (Collins Primary Dictionaries)
0
id-0008128006
None
A MIND OF YOUR OWN THE TRUTH A
0
id-000820571X
None
Greek Audio Course (Collins Easy Learning Audio Course)
0
id-000820666X
None
A Postcard from Italy: The most uplifting and escapist romance for 2023 from the No.1 bestseller (Book 1)
0
id-0008215995
None
Bloodchild
0
id-000824121X
None
The Times Fiendish Su Doku Book 11: 200 Challenging Su Doku Puzzles
0
id-0008359202
None
Emperors of the Deep: The Mysterious and Misunderstood World of the Shark
0
id-0008495742
None
The Guilty Couple: The completely nail-biting, unputdownable crime thriller from the international million-copy bestseller
0
id-0008527229
None
BOSH! on a Budget: From the bestselling vegan authors comes the latest healthy plant-based, meat-free cookbook with new deliciously simple recipes
0
id-002035441X
None
Ordeal of the Union Vol.1: Fruits of Manifest Destiny 1847-1852 : A House Dividing 1852-1857
0
id-0020354436
None
Ordeal of the Union Vol. 3: The Improvised War 1861-1862; War Becomes Revolution 1862-1863
0
id-0020625804
None
Birds of North America: Western Region : A Quick Identification Guide for All Bird-Watchers (Macmillan Field Guides)
0
id-0021150214
None
McGraw-Hill My Math, Grade 2, Student Edition, Volume 1 (ELEMENTARY MATH CONNECTS)
0
id-0023611367
None
C for Scientists and Engineers
0
id-0023795204
None
Writing of Economics
0
id-0023882204
None
Writing Exercises: Building, Combining, and Revising
0
id-0024035459
None
Approaches to Early Childhood Education
0
id-0024174009
None
Theory of Price
0
id-0025074903
None
Royal Service: My Twelve Years As Valet to Prince Charles
0
id-0026123002
None
Liquid Assets: How to Develop an Enjoyable and Profitable Wine Portfolio
0
id-0028619897
None
The Pro Football Encyclopedia: The Complete and Definitive Record of Professional Football
0
id-0028720105
None
Straight Life: The Story of Art Pepper
0
id-0029183308
None
Power and Influence
0
id-0029268214
None
Lives on the Boundary
0
id-0030210291
None
PKG:FINL MGMT 9E+STD SPDSHT APP DSK (Dryden Press Series in Finance)
0
id-0030384427
None
A Book of Short Stories 1
0
id-0030914566
None
Soldier
0
id-0046410120
None
The art of vegetarian cookery
0
id-0060000023
None
Warriors: Into the Wild
0
id-0060005556
None
Last to Die (Jack Swyteck Novel)
0
id-0060083700
None
Rooms: Creating Luxurious, Livable Spaces (Design)
0
id-0060083913
None
How to Cook Revised Edition: An Easy and Imaginative Guide for the Beginner
0
id-0060087919
None
Shockproof Sydney Skate
0
id-0060142774
None
The Cuban Revolution
0
id-0060147326
None
No Chinese Stranger (1st Edition)
0
id-0060154551
None
A Christmas Treasury
0
id-0060156783
None
Make Way For Lucia: The Complete Lucia, Including Queen Lucia / Lucia in London / Miss Mapp / The Male Impersonator / Mapp and Lucia / The Worshipful Lucia / Trouble for Lucia
0
id-0060199733
None
Statecraft: Strategies for a Changing World
0
id-0060236035
None
The Rooster's Gift
0
id-0060241020
None
Sarah, Plain and Tall: A Newbery Award Winner (Sarah, Plain and Tall, 1)
0
id-0060241039
None
Back home
0
id-0060263458
None
Cowboys and Cattle Country (American Heritage Junior Library)
0
id-006039157X
None
Think a Second Time
0
id-0060439785
None
Music: A Listener's Introduction
0
id-0060502266
None
A Game of Scandal
0
id-0060538252
None
She Comes First: The Thinking Man's Guide to Pleasuring a Woman
0
id-0060560177
None
Backstage Pass
0
id-0060582723
None
What's Right with Islam: A New Vision for Muslims and the West
0
id-006058338X
None
The Berenstain Bears Play T-Ball (I Can Read Level 1)
0
id-0060595353
None
Homecourt Advantage
0
id-0060621516
None
The HarperCollins Concise Guide to World Religion: The A-to-Z Encyclopedia of All the Major Religious Traditions
0
id-0060652578
None
A man under orders: Lieutenant General William K. Harrison, Jr
0
id-006075978X
None
Twins: A Novel
0
id-006077620X
None
Sugar Rush
0
id-0060783109
None
The Dog Princess Fairy Tails
0
id-006079335X
None
Prince Caspian (The Chronicles of Narnia)
0
id-006082221X
None
Annette Vallon: A Novel of the French Revolution
0
id-006082512X
None
The Big Book of Women Saints
0
id-006083353X
None
The Loathsome Library: A Box of Unfortunate Events, Books 1-6 (The Bad Beginning; The Reptile Room; The Wide Window; The Miserable Mill; The Austere Academy; The Ersatz Elevator)
0
id-0060846747
None
Sex with the Queen: 900 Years of Vile Kings, Virile Lovers, and Passionate Politics (P.S.)
0
id-0060846844
None
Lost Girls and Love Hotels: A Novel
0
id-0060851988
None
Queen of Babble: A Novel
0
id-006090691X
None
Self-exposures: A workbook in photographic self-portraiture
0
id-0060910607
None
Writing With a Word Processor
0
id-0060924691
None
Into The Garden: A Wedding Anthology: Poetry and Prose on Love and Marriage
0
id-0060929103
None
Absolute Beauty: Radiant Skin and Inner Harmony Through the Ancient Secrets of Ayurveda
0
id-0060973293
None
Market Wizards: Interviews with Top Traders
0
id-0061004251
None
Crazymaker: The Shocking True Story of Murder and Betrayal in an American Family
0
id-0061053082
None
Tapestries: An Anthology (Magic : The Gathering)
0
id-0061080985
None
Heart Sounds (Harper Monogram)
0
id-006115430X
None
Universe of Stone: Chartres Cathedral and the Invention of the Gothic AKA Universe of Stone: A Biography of Chartres Cathedral
0
id-0061189138
None
El Ticket de Tu Vida
0
id-0061251445
None
Smithsonian Atlas of World Aviation
0
id-0061256412
None
Michael Tolliver Lives CD
0
id-0061288845
None
The Coffin Club (Vampire Kisses, Book 5)
0
id-0061363936
None
Run: A Novel
0
id-0061379093
None
Ana's Story: A Journey of Hope
0
id-0061450650
None
The Dress Doctor: Prescriptions for Style, From A to Z
0
id-0061473723
None
Zack's Alligator and the First Snow: A Winter and Holiday Book for Kids (I Can Read Level 2)
0
id-0061478253
None
Little Critter 12-Book Phonics Fun!: Includes 12 Mini-Books Featuring Short and Long Vowel Sounds (My First I Can Read)
0
id-0061624810
None
The Map of True Places: A Novel
0
id-0061697214
None
Broker, Trader, Lawyer, Spy: The Secret World of Corporate Espionage
0

Marqo Ecommerce Embedding Models

In this work, we introduce the GoogleShopping-1m dataset for evaluation. This dataset comes with the release of our state-of-the-art embedding models for ecommerce products: Marqo-Ecommerce-B and Marqo-Ecommerce-L.

Released Content:

  1. Marqo-Ecommerce-B and Marqo-Ecommerce-L embedding models
  2. GoogleShopping-1m and AmazonProducts-3m for evaluation
  3. Evaluation Code

The benchmarking results show that the Marqo-Ecommerce models consistently outperformed all other models across various metrics. Specifically, marqo-ecommerce-L achieved an average improvement of 17.6% in MRR and 20.5% in nDCG@10 when compared with the current best open source model, ViT-SO400M-14-SigLIP across all three tasks in the marqo-ecommerce-hard dataset. When compared with the best private model, Amazon-Titan-Multimodal, we saw an average improvement of 38.9% in MRR and 45.1% in nDCG@10 across all three tasks, and 35.9% in Recall across the Text-to-Image tasks in the marqo-ecommerce-hard dataset.

multi split visual

More benchmarking results can be found below.

Models

Embedding Model #Params (m) Dimension HuggingFace Download .pt
Marqo-Ecommerce-B 203 768 Marqo/marqo-ecommerce-embeddings-B link
Marqo-Ecommerce-L 652 1024 Marqo/marqo-ecommerce-embeddings-L link

Load from HuggingFace with transformers

To load the models in Transformers, see below. The models are hosted on Hugging Face and loaded using Transformers.

from transformers import AutoModel, AutoProcessor
import torch
from PIL import Image
import requests

model_name= 'Marqo/marqo-ecommerce-embeddings-L'
# model_name = 'Marqo/marqo-ecommerce-embeddings-B'

model = AutoModel.from_pretrained(model_name, trust_remote_code=True)
processor = AutoProcessor.from_pretrained(model_name, trust_remote_code=True)

img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-ecommerce-embeddings/refs/heads/main/images/dining-chairs.png', stream=True).raw).convert("RGB")
image = [img]
text = ["dining chairs", "a laptop", "toothbrushes"]
processed = processor(text=text, images=image, padding='max_length', return_tensors="pt")
processor.image_processor.do_rescale = False
with torch.no_grad():
    image_features = model.get_image_features(processed['pixel_values'], normalize=True)
    text_features = model.get_text_features(processed['input_ids'], normalize=True)

    text_probs = (100 * image_features @ text_features.T).softmax(dim=-1)
    
print(text_probs)
# [1.0000e+00, 8.3131e-12, 5.2173e-12]

Load from HuggingFace with OpenCLIP

To load the models in OpenCLIP, see below. The models are hosted on Hugging Face and loaded using OpenCLIP. You can also find this code inside run_models.py.

pip install open_clip_torch
from PIL import Image
import open_clip
import requests
import torch

# Specify model from Hugging Face Hub
model_name = 'hf-hub:Marqo/marqo-ecommerce-embeddings-L'
# model_name = 'hf-hub:Marqo/marqo-ecommerce-embeddings-B'

model, preprocess_train, preprocess_val = open_clip.create_model_and_transforms(model_name)
tokenizer = open_clip.get_tokenizer(model_name)

# Preprocess the image and tokenize text inputs
# Load an example image from a URL
img = Image.open(requests.get('https://raw.githubusercontent.com/marqo-ai/marqo-ecommerce-embeddings/refs/heads/main/images/dining-chairs.png', stream=True).raw)
image = preprocess_val(img).unsqueeze(0)
text = tokenizer(["dining chairs", "a laptop", "toothbrushes"])

# Perform inference
with torch.no_grad(), torch.cuda.amp.autocast():
    image_features = model.encode_image(image, normalize=True)
    text_features = model.encode_text(text, normalize=True)

    # Calculate similarity probabilities
    text_probs = (100.0 * image_features @ text_features.T).softmax(dim=-1)

# Display the label probabilities
print("Label probs:", text_probs)
# [1.0000e+00, 8.3131e-12, 5.2173e-12]

Evaluation

Generalised Contrastiove Learning (GCL) is used for the evaluation. The following code can also be found in scripts.

git clone https://github.com/marqo-ai/GCL

Install the packages required by GCL.

1. GoogleShopping-Text2Image Retrieval.

cd ./GCL
MODEL=hf-hub:Marqo/marqo-ecommerce-B
outdir=/MarqoModels/GE/marqo-ecommerce-B/gs-title2image
hfdataset=Marqo/google-shopping-general-eval
python  evals/eval_hf_datasets_v1.py \
      --model_name $MODEL \
      --hf-dataset $hfdataset \
      --output-dir $outdir \
      --batch-size 1024 \
      --num_workers 8 \
      --left-key "['title']" \
      --right-key "['image']" \
      --img-or-txt "[['txt'], ['img']]" \
      --left-weight "[1]" \
      --right-weight "[1]" \
      --run-queries-cpu \
      --top-q 4000 \
      --doc-id-key item_ID \
      --context-length "[[64], [0]]"

2. GoogleShopping-Category2Image Retrieval.

cd ./GCL
MODEL=hf-hub:Marqo/marqo-ecommerce-B
outdir=/MarqoModels/GE/marqo-ecommerce-B/gs-cat2image
hfdataset=Marqo/google-shopping-general-eval
python  evals/eval_hf_datasets_v1.py \
      --model_name $MODEL \
      --hf-dataset $hfdataset \
      --output-dir $outdir \
      --batch-size 1024 \
      --num_workers 8 \
      --left-key "['query']" \
      --right-key "['image']" \
      --img-or-txt "[['txt'], ['img']]" \
      --left-weight "[1]" \
      --right-weight "[1]" \
      --run-queries-cpu \
      --top-q 4000 \
      --doc-id-key item_ID \
      --context-length "[[64], [0]]"

3. AmazonProducts-Category2Image Retrieval.

cd ./GCL
MODEL=hf-hub:Marqo/marqo-ecommerce-B
outdir=/MarqoModels/GE/marqo-ecommerce-B/ap-title2image
hfdataset=Marqo/amazon-products-eval
python  evals/eval_hf_datasets_v1.py \
      --model_name $MODEL \
      --hf-dataset $hfdataset \
      --output-dir $outdir \
      --batch-size 1024 \
      --num_workers 8 \
      --left-key "['title']" \
      --right-key "['image']" \
      --img-or-txt "[['txt'], ['img']]" \
      --left-weight "[1]" \
      --right-weight "[1]" \
      --run-queries-cpu \
      --top-q 4000 \
      --doc-id-key item_ID \
      --context-length "[[64], [0]]"

Detailed Performance

Our benchmarking process was divided into two distinct regimes, each using different datasets of ecommerce product listings: marqo-ecommerce-hard and marqo-ecommerce-easy. Both datasets contained product images and text and only differed in size. The "easy" dataset is approximately 10-30 times smaller (200k vs 4M products), and designed to accommodate rate-limited models, specifically Cohere-Embeddings-v3 and GCP-Vertex (with limits of 0.66 rps and 2 rps respectively). The "hard" dataset represents the true challenge, since it contains four million ecommerce product listings and is more representative of real-world ecommerce search scenarios.

Within both these scenarios, the models were benchmarked against three different tasks:

  • Google Shopping Text-to-Image
  • Google Shopping Category-to-Image
  • Amazon Products Text-to-Image

Marqo-Ecommerce-Hard

Marqo-Ecommerce-Hard looks into the comprehensive evaluation conducted using the full 4 million dataset, highlighting the robust performance of our models in a real-world context.

GoogleShopping-Text2Image Retrieval.

Embedding Model mAP R@10 MRR nDCG@10
Marqo-Ecommerce-L 0.682 0.878 0.683 0.726
Marqo-Ecommerce-B 0.623 0.832 0.624 0.668
ViT-SO400M-14-SigLip 0.573 0.763 0.574 0.613
ViT-L-16-SigLip 0.540 0.722 0.540 0.577
ViT-B-16-SigLip 0.476 0.660 0.477 0.513
Amazon-Titan-MultiModal 0.475 0.648 0.475 0.509
Jina-V1-CLIP 0.285 0.402 0.285 0.306

GoogleShopping-Category2Image Retrieval.

Embedding Model mAP P@10 MRR nDCG@10
Marqo-Ecommerce-L 0.463 0.652 0.822 0.666
Marqo-Ecommerce-B 0.423 0.629 0.810 0.644
ViT-SO400M-14-SigLip 0.352 0.516 0.707 0.529
ViT-L-16-SigLip 0.324 0.497 0.687 0.509
ViT-B-16-SigLip 0.277 0.458 0.660 0.473
Amazon-Titan-MultiModal 0.246 0.429 0.642 0.446
Jina-V1-CLIP 0.123 0.275 0.504 0.294

AmazonProducts-Text2Image Retrieval.

Embedding Model mAP R@10 MRR nDCG@10
Marqo-Ecommerce-L 0.658 0.854 0.663 0.703
Marqo-Ecommerce-B 0.592 0.795 0.597 0.637
ViT-SO400M-14-SigLip 0.560 0.742 0.564 0.599
ViT-L-16-SigLip 0.544 0.715 0.548 0.580
ViT-B-16-SigLip 0.480 0.650 0.484 0.515
Amazon-Titan-MultiModal 0.456 0.627 0.457 0.491
Jina-V1-CLIP 0.265 0.378 0.266 0.285

Marqo-Ecommerce-Easy

This dataset is about 10-30 times smaller than the Marqo-Ecommerce-Hard, and designed to accommodate rate-limited models, specifically Cohere-Embeddings-v3 and GCP-Vertex.

GoogleShopping-Text2Image Retrieval.

Embedding Model mAP R@10 MRR nDCG@10
Marqo-Ecommerce-L 0.879 0.971 0.879 0.901
Marqo-Ecommerce-B 0.842 0.961 0.842 0.871
ViT-SO400M-14-SigLip 0.792 0.935 0.792 0.825
GCP-Vertex 0.740 0.910 0.740 0.779
ViT-L-16-SigLip 0.754 0.907 0.754 0.789
ViT-B-16-SigLip 0.701 0.870 0.701 0.739
Amazon-Titan-MultiModal 0.694 0.868 0.693 0.733
Jina-V1-CLIP 0.480 0.638 0.480 0.511
Cohere-embedding-v3 0.358 0.515 0.358 0.389

GoogleShopping-Category2Image Retrieval.

Embedding Model mAP P@10 MRR nDCG@10
Marqo-Ecommerce-L 0.515 0.358 0.764 0.590
Marqo-Ecommerce-B 0.479 0.336 0.744 0.558
ViT-SO400M-14-SigLip 0.423 0.302 0.644 0.487
GCP-Vertex 0.417 0.298 0.636 0.481
ViT-L-16-SigLip 0.392 0.281 0.627 0.458
ViT-B-16-SigLip 0.347 0.252 0.594 0.414
Amazon-Titan-MultiModal 0.308 0.231 0.558 0.377
Jina-V1-CLIP 0.175 0.122 0.369 0.229
Cohere-embedding-v3 0.136 0.110 0.315 0.178

AmazonProducts-Text2Image Retrieval.

Embedding Model mAP R@10 MRR nDCG@10
Marqo-Ecommerce-L 0.92 0.978 0.928 0.940
Marqo-Ecommerce-B 0.897 0.967 0.897 0.914
ViT-SO400M-14-SigLip 0.860 0.954 0.860 0.882
ViT-L-16-SigLip 0.842 0.940 0.842 0.865
GCP-Vertex 0.808 0.933 0.808 0.837
ViT-B-16-SigLip 0.797 0.917 0.797 0.825
Amazon-Titan-MultiModal 0.762 0.889 0.763 0.791
Jina-V1-CLIP 0.530 0.699 0.530 0.565
Cohere-embedding-v3 0.433 0.597 0.433 0.465

Citation

@software{zhu2024marqoecommembed_2024,
        author = {Tianyu Zhu and and Jesse Clark},
        month = oct,
        title = {{Marqo Ecommerce Embeddings - Foundation Model for Product Embeddings}},
        url = {https://github.com/marqo-ai/marqo-ecommerce-embeddings/},
        version = {1.0.0},
        year = {2024}
        }
Downloads last month
94

Collection including Marqo/amazon-products-eval-100k