RECORD
Understanding and Comparing Deep Neural Networks for Age and Gender Classification
Machine Learning · Submitted 25 Aug 2017
abstract compiled · 2026-09-23 · no language-model review
The authors' main claim With a combination of simple preprocessing steps, we reach state of the art performance in gender recognition.
The compiled record
The paper's own sentences, arranged. Each label is drawn from a closed vocabulary; the number beside it is the probability the classifier gave that label. Numbers in the tables are the authors', checked by code. No sentence in this record was written by a model.
Open the compiled record
The abstract, sentence by sentence
Recently, deep neural networks have demonstrated excellent performances in recognizing the age and gender on human face images. However, these models were applied in a black-box manner with no information provided about which facial features are actually used for prediction and how these features depend on image preprocessing, model initialization and architecture choice. We present a study investigating these different effects. In detail, our work compares four popular neural network architectures, studies the effect of pretraining, evaluates the robustness of the considered alignment preprocessings via cross-method test set swapping and intuitively visualizes the model's prediction strategies in given preprocessing conditions using the recent Layer-wise Relevance Propagation (LRP) algorithm. Our evaluations on the challenging Adience benchmark show that suitable parameter initialization leads to a holistic perception of the input, compensating artefactual data representations. With a combination of simple preprocessing steps, we reach state of the art performance in gender recognition.
background contribution method result interpretation main claim
| # | sentence | role | the authors' own work | states the main conclusion | strength |
|---|---|---|---|---|---|
| s0 | Recently, deep neural networks have demonstrated excellent performances in recognizing the age and gender on human face images. | background 1.00 | 0.05 | 0.03 | asserted 1.00 |
| s1 | However, these models were applied in a black-box manner with no information provided about which facial features are actually used for prediction and how these features depend on image preprocessing, model initialization and architecture choice. | gap 1.00 | 0.07 | 0.02 | asserted 1.00 |
| s2 | We present a study investigating these different effects. | contribution 1.00 | 0.72 | 0.03 | asserted 0.58 |
| s3 | In detail, our work compares four popular neural network architectures, studies the effect of pretraining, evaluates the robustness of the considered alignment preprocessings via cross-method test set swapping and intuitively visualizes the model's prediction strategies in given preprocessing conditions using the recent Layer-wise Relevance Propagation (LRP) algorithm. | method 0.69 | 0.95 | 0.06 | asserted 0.89 |
| s4 | Our evaluations on the challenging Adience benchmark show that suitable parameter initialization leads to a holistic perception of the input, compensating artefactual data representations. | result 0.73 | 0.95 | 0.33 | asserted 1.00 |
| s5 | With a combination of simple preprocessing steps, we reach state of the art performance in gender recognition. | result 1.00 | 0.94 | 0.31 | asserted 1.00 |
What was examined, and what was not
- arXiv listing abstract v1
- full text
- figures and tables
- references and citing sentences
- relations to other papers
- P abstract
- S sections
- T body sentences
- C citations
- R cross-paper relations
The paper
Original abstract
Recently, deep neural networks have demonstrated excellent performances in recognizing the age and gender on human face images. However, these models were applied in a black-box manner with no information provided about which facial features are actually used for prediction and how these features depend on image preprocessing, model initialization and architecture choice. We present a study investigating these different effects. In detail, our work compares four popular neural network architectures, studies the effect of pretraining, evaluates the robustness of the considered alignment preprocessings via cross-method test set swapping and intuitively visualizes the model's prediction strategies in given preprocessing conditions using the recent Layer-wise Relevance Propagation (LRP) algorithm. Our evaluations on the challenging Adience benchmark show that suitable parameter initialization leads to a holistic perception of the input, compensating artefactual data representations. With a combination of simple preprocessing steps, we reach state of the art performance in gender recognition.
Human review
How the literature uses it
2 arXiv papers cite this in 3 sentences
Papers citing it · 2 sentences without a confident label
Uses its method or tool 1 paper
-
Interpretability Methods We compared three different interpretability methods for the EmoPy computer vision model, a) the gradient of output w.r.t. the input image, b) Layerwise relevance propagation (lrp) [ref] and c) Guided backpropagation [ref] .
Quality Metrics for Transparent Machine Learning With and Without Humans In the Loop Are Not Correlated · Experiments · 0.96
What it cites · 16 sentences without a confident label
Contests 1 paper
-
We rectified the performance of [ref] on gender recognition referred to in [ref] with a more likely result and report our own result, slightly exceeding that baseline.
DAGER: Deep Age, Gender and Emotion Recognition Using Convolutional Neural Network · Introduction · contest 0.79
Uses the method or tool of 1 paper
-
All models are trained using the Caffe Deep Learning Framework [ref] , with code based on <https://github.com/GilLevi/AgeGenderDeepLearning>, containing the configurations to reproduce the results from [ref] .
Caffe: Convolutional Architecture for Fast Feature Embedding · Architectures, Preprocessing and Model Initialization · 1.00
Cites as background 5 papers
-
Improvements in network architecture and model performance have been steady and fast-paced since then [ref] .
Inception-v4, Inception-ResNet and the Impact of Residual Connections on Learning · Introduction · 0.94
-
Improvements in network architecture and model performance have been steady and fast-paced since then [ref] .
Very Deep Convolutional Networks for Large-Scale Image Recognition · Introduction · 0.94
-
Several methods for explaining individual predictions have emerged since then, with robust yet computationally expensive occlusion-based [ref] and sampling-based analysis [ref] , (gradient-based) sensitivity analysis [ref] and backpropagation-type approaches [ref] among them.
A New Method to Visualize Deep Neural Networks · Visualizing Model Perception · 0.97
Discussion
Know something about this paper? Leave a comment, or highlight any passage above to comment on it.