Fragment embedding

Closed weights Stanford University 144.5M parameters June 2014

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
Stanford University
Organisation type
Academia
Country
United States of America
Published
21 June 2014
Authors
A. Karpathy, Armand Joulin, Li Fei-Fei

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Vision
Task
Entity embedding, Semantic segmentation

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
144.5M

Model contains a word embedding. a matrix combining two word embeddings, and image embedding (built upon a pretrained RCNN image model. Word embedding: 400000 * 200 =80000000 ("Here, We is a d × 400, 000 matrix that encodes a 1-of-k vector into a d-dimensional word vector representation (we use d = 200)." Embedding dimension: 1000 ("The size of the embedded space is cross-validated, and we found that values of approximately 1000 generally work well." Word combination matrix: 400* 1000=400000 Ima…

Training data
15,000,000 tokens

Largest experiment uses 30000 training images, 30000 * 5 = 150,000 sentences

Epochs
20

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Why it is tracked
SOTA improvement

"Extensive experimental evaluation shows that reasoning on both the global level of images and sentences and the finer level of their respective fragments significantly improves performance on image-sentence retrieval tasks."

Record confidence
Likely

Sources

Where this record came from and when it was last checked.

Reference
Deep Fragment Embeddings for Bidirectional Image Sentence Mapping
Last updated
28 November 2025

What the numbers mean

Where it came from

Fragment embedding was published by Stanford University, in United States of America, in June 2014. It comes out of academia.

It works in Vision, and is recorded as doing entity embedding, Semantic segmentation.

Because the weights are not available, none of the hardware figures elsewhere on this site apply to it.

What went into building it

The training set ran to roughly 15,000,000 tokens.

It is tracked in the underlying dataset for one reason in particular: sOTA improvement.

Answers

Fragment embedding — common questions

01

How many parameters does Fragment embedding have?

Fragment embedding has 144.5M parameters. Model contains a word embedding. a matrix combining two word embeddings, and image embedding (built upon a pretrained RCNN image model. Word embedding: 400000 * 200 =80000000 ("Here, We is a d × 400, 000 matrix that encodes a 1-of-k vector into a d-dimensional word vector representation (we use d = 200)." Embedding dimension: 1000 ("The size of the embedded space is cross-validated, and we found that values of approximately 1000 generally work well." Word combination matrix: 400* 1000=400000 Image embedding: 4096*1000=4096000 ("We use the Caffe [41] implementation of the ImageNet Detection RCNN model [27] to detect objects in all images. On our machine with a Tesla K40 GPU, the RCNN processes one image in approximately 25 seconds. We discard the predictions for 200 ImageNet detection classes and only keep the 4096-D activations") CNN: 60,000,000 "The CNN architecture is identical to the one described in Girhsick et al. [26]. It contains approximately 60 million parameters" Total parameters: 4096000+80000000+400000+60000000=144,496,000. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

02

Who created Fragment embedding?

Fragment embedding was published by Stanford University, based in United States of America, categorised as academia.

03

When was Fragment embedding released?

Fragment embedding was published in June 2014. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

04

What is Fragment embedding used for?

Fragment embedding works in Vision, and is recorded as handling entity embedding, Semantic segmentation. A model can carry several of each, so these are the areas it was built for rather than a limit on what it will attempt.

05

What GPU do I need to run Fragment embedding?

None. Fragment embedding is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

06

Is Fragment embedding open source?

The licensing for Fragment embedding was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.