DeiT
E435881
DeiT is a family of data-efficient vision transformer models designed for image classification with reduced training data requirements and strong performance.
All labels observed (1)
| Label | Occurrences |
|---|---|
| DeiT canonical | 1 |
How this entity was disambiguated
This entity first appeared as the object of triple T4389209 — resolving that mention is where its identity was fixed. The disambiguator weighed these candidate entities and picked the highlighted one (or “None”, minting a new entity). This is how homonymy is resolved: the same surface form can point to different entities.
NED1
Entity disambiguation (via context triple)
gpt-5-mini-2025-08-07
Target entity: DeiT Context triple: [Hugging Face Transformers, supportsModelType, DeiT]
-
A.
ResNeXt
ResNeXt is a deep convolutional neural network architecture that extends ResNet by using grouped convolutions and a split-transform-merge strategy to improve accuracy and efficiency in image recognition tasks.
-
B.
DETR
DETR is the acronym for the former UK government Department of the Environment, Transport and the Regions, which was responsible for environmental policy, transport, and regional affairs.
-
C.
ResNet
ResNet is a deep convolutional neural network architecture known for its use of residual connections to enable very deep models and achieve state-of-the-art performance in image recognition tasks.
-
D.
PWSFTviT
PWSFTviT is the renowned Łódź Film School in Poland, one of Europe’s leading film and television academies known for training many acclaimed filmmakers.
-
E.
CLIP
CLIP is an OpenAI model that learns joint representations of images and text, enabling tasks like zero-shot image classification and natural language-based image retrieval.
- F. None of above. chosen
- G. Unsure - the case is ambiguous/there is not enough information to decide.
NED2
Entity disambiguation (via description)
gpt-5-mini-2025-08-07
Target entity: DeiT Target entity description: DeiT is a family of data-efficient vision transformer models designed for image classification with reduced training data requirements and strong performance.
-
A.
ResNeXt
ResNeXt is a deep convolutional neural network architecture that extends ResNet by using grouped convolutions and a split-transform-merge strategy to improve accuracy and efficiency in image recognition tasks.
-
B.
DETR
DETR is the acronym for the former UK government Department of the Environment, Transport and the Regions, which was responsible for environmental policy, transport, and regional affairs.
-
C.
ResNet
ResNet is a deep convolutional neural network architecture known for its use of residual connections to enable very deep models and achieve state-of-the-art performance in image recognition tasks.
-
D.
PWSFTviT
PWSFTviT is the renowned Łódź Film School in Poland, one of Europe’s leading film and television academies known for training many acclaimed filmmakers.
-
E.
CLIP
CLIP is an OpenAI model that learns joint representations of images and text, enabling tasks like zero-shot image classification and natural language-based image retrieval.
- F. None of above. chosen
Statements (48)
| Predicate | Object |
|---|---|
| instanceOf |
deep learning model
ⓘ
image classification model ⓘ vision transformer model family ⓘ |
| abbreviationFor | Data-efficient Image Transformers NERFINISHED ⓘ |
| availableIn |
Hugging Face Transformers
NERFINISHED
ⓘ
timm library NERFINISHED ⓘ |
| basedOn |
ViT architecture
ⓘ
Vision Transformer NERFINISHED ⓘ |
| codeRepository | https://github.com/facebookresearch/deit ⓘ |
| contribution |
reduced data requirements for vision transformers
ⓘ
showed transformers can be trained from scratch on ImageNet-1k only ⓘ |
| designedFor | image classification ⓘ |
| developedBy |
Facebook AI Research
NERFINISHED
ⓘ
Meta AI NERFINISHED ⓘ |
| doesNotRequire |
ImageNet-21k pretraining
ⓘ
JFT pretraining ⓘ |
| firstAuthor | Hugo Touvron NERFINISHED ⓘ |
| fullName | Data-efficient Image Transformers NERFINISHED ⓘ |
| hasAuthor |
Alexandre Sablayrolles
NERFINISHED
ⓘ
Gabriel Synnaeve NERFINISHED ⓘ Herve Jegou NERFINISHED ⓘ Hugo Touvron NERFINISHED ⓘ Matthieu Cord NERFINISHED ⓘ |
| hasProperty |
competitive accuracy on ImageNet
ⓘ
reduced training data requirements ⓘ strong performance with limited data ⓘ |
| hasVariant |
DeiT-B
NERFINISHED
ⓘ
DeiT-B distilled NERFINISHED ⓘ DeiT-S NERFINISHED ⓘ DeiT-S distilled NERFINISHED ⓘ DeiT-Ti NERFINISHED ⓘ DeiT-Ti distilled NERFINISHED ⓘ |
| implementedIn | PyTorch NERFINISHED ⓘ |
| inputType | 2D images ⓘ |
| introducedDistillationMethod | token-based knowledge distillation ⓘ |
| license | Apache-2.0 (for official code release) ⓘ |
| optimizedFor | data efficiency ⓘ |
| paperTitle | Training data-efficient image transformers & distillation through attention NERFINISHED ⓘ |
| publicationYear | 2020 ⓘ |
| publishedAs | research paper ⓘ |
| task | supervised image classification ⓘ |
| teacherModelType | convolutional neural network ⓘ |
| trainingDataset | ImageNet-1k NERFINISHED ⓘ |
| uses |
class token
ⓘ
distillation token ⓘ knowledge distillation from CNN teacher ⓘ patch embedding ⓘ transformer encoder ⓘ |
How these facts were elicited
The pipeline generated the facts above by prompting gpt-5.1 with this entity's name + description and the instruction below.
Instruction
You are a knowledge base construction expert. Given a subject entity and a description of it, return factual statements that you know for the subject as a JSON list of dictionaries(triples), where keys must be "subject", "predicate" and "object". The number of facts may be very high, between 25 to 50 or more, for very popular subjects. For less popular subjects, the number of facts can be very low, like 5 or 10. # Requirements - If you don't know the subject at all, return an empty list. - If the subject is not a named entity, return an empty list. - Include at least one triple where predicate is "instanceOf". - Do not get too wordy. - Separate several objects into multiple triples with one object.
Input
Subject: DeiT Description of subject: DeiT is a family of data-efficient vision transformer models designed for image classification with reduced training data requirements and strong performance.
Referenced by (1)
Full triples — surface form annotated when it differs from this entity's canonical label.