Inception Score
E290873
Inception Score is a quantitative metric used to assess the quality and diversity of images generated by generative models by analyzing their classifiability and distribution across categories using a pretrained Inception network.
All labels observed (1)
| Label | Occurrences |
|---|---|
| Inception Score canonical | 4 |
How this entity was disambiguated
This entity first appeared as the object of triple T2703886 — 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: Inception Score Context triple: [Generative Adversarial Networks, evaluationMetric, Inception Score]
-
A.
Kullback–Leibler divergence
Kullback–Leibler divergence is a fundamental information-theoretic measure that quantifies how one probability distribution differs from a reference distribution.
-
B.
Rényi entropy
Rényi entropy is a generalized measure of information and uncertainty that extends Shannon entropy by introducing a tunable order parameter to emphasize different aspects of a probability distribution.
-
C.
Shannon entropy
Shannon entropy is a fundamental measure in information theory that quantifies the average uncertainty or information content in a random variable or message source.
-
D.
Bhattacharyya distance
Bhattacharyya distance is a statistical measure of similarity between two probability distributions, often used in pattern recognition and classification to quantify their overlap.
-
E.
Tsallis divergence
Tsallis divergence is a generalized measure of statistical distance between probability distributions derived from Tsallis entropy, often used in nonextensive statistical mechanics and information theory.
- 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: Inception Score Target entity description: Inception Score is a quantitative metric used to assess the quality and diversity of images generated by generative models by analyzing their classifiability and distribution across categories using a pretrained Inception network.
-
A.
Kullback–Leibler divergence
Kullback–Leibler divergence is a fundamental information-theoretic measure that quantifies how one probability distribution differs from a reference distribution.
-
B.
Rényi entropy
Rényi entropy is a generalized measure of information and uncertainty that extends Shannon entropy by introducing a tunable order parameter to emphasize different aspects of a probability distribution.
-
C.
Shannon entropy
Shannon entropy is a fundamental measure in information theory that quantifies the average uncertainty or information content in a random variable or message source.
-
D.
Bhattacharyya distance
Bhattacharyya distance is a statistical measure of similarity between two probability distributions, often used in pattern recognition and classification to quantify their overlap.
-
E.
Tsallis divergence
Tsallis divergence is a generalized measure of statistical distance between probability distributions derived from Tsallis entropy, often used in nonextensive statistical mechanics and information theory.
- F. None of above. chosen
Statements (48)
| Predicate | Object |
|---|---|
| instanceOf |
evaluation metric
ⓘ
image generation quality metric ⓘ quantitative metric ⓘ |
| alternativeTo | Fréchet Inception Distance ⓘ |
| appliesTo |
GAN-generated images
ⓘ
image synthesis models ⓘ images generated by generative models ⓘ |
| assumes |
diverse image sets have high-entropy marginal label distribution
ⓘ
high-quality images have low-entropy label distributions ⓘ |
| basedOn |
Inception architecture
ⓘ
surface form:
Inception network
pretrained Inception v3 classifier ⓘ |
| category |
computer vision metric
ⓘ
machine learning metric ⓘ |
| commonlyComputedOn |
CIFAR-10
ⓘ
surface form:
CIFAR-10 dataset
ImageNet ⓘ
surface form:
ImageNet-like datasets
|
| comparedWith | Fréchet Inception Distance ⓘ |
| definedAs | exponential of expected KL divergence between p(y|x) and p(y) ⓘ |
| dependsOn |
choice of pretrained Inception model
ⓘ
dataset used to train Inception network ⓘ |
| domain | deep generative modeling ⓘ |
| hasFormula | IS = exp( E_x[ KL( p(y|x) || p(y) ) ] ) ⓘ |
| higherIs | better ⓘ |
| implementedIn | popular deep learning libraries and toolkits ⓘ |
| introducedBy |
Ian Goodfellow
ⓘ
Tim Salimans ⓘ |
| introducedInContextOf | Generative Adversarial Networks ⓘ |
| introducedInPaper | Improved Techniques for Training GANs ⓘ |
| introducedInYear | 2016 ⓘ |
| limitation |
can be gamed by overfitting to Inception classifier
ⓘ
does not compare to real data distribution directly ⓘ not well correlated with human perceptual quality in all settings ⓘ sensitive to mode dropping ⓘ |
| measures |
KL divergence between conditional and marginal label distributions
ⓘ
classifiability of generated images ⓘ diversity across predicted classes ⓘ |
| relatedConcept |
image diversity
ⓘ
image realism ⓘ mode collapse ⓘ |
| requires |
fixed pretrained classifier
ⓘ
large set of generated images ⓘ |
| usedFor |
assessing diversity of generated images
ⓘ
assessing quality of generated images ⓘ benchmarking image generative models ⓘ evaluating generative models ⓘ |
| usedIn |
GAN research literature
ⓘ
evaluation of image-to-image translation models ⓘ evaluation of unconditional image generation ⓘ |
| uses | softmax output of Inception network ⓘ |
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: Inception Score Description of subject: Inception Score is a quantitative metric used to assess the quality and diversity of images generated by generative models by analyzing their classifiability and distribution across categories using a pretrained Inception network.
Referenced by (4)
Full triples — surface form annotated when it differs from this entity's canonical label.