pith. sign in

arxiv: 2504.15594 · v1 · submitted 2025-04-22 · 💻 cs.LG · cs.CV

Analytical Softmax Temperature Setting from Feature Dimensions for Model- and Domain-Robust Classification

Tatsuhito Hasegawa , Shunsuke Sakai This is my paper

Pith reviewed 2026-05-22 18:35 UTC · model grok-4.3

classification 💻 cs.LG cs.CV
keywords softmax temperaturefeature dimensionalitytraining-freeclassificationbatch normalizationempirical formulatemperature scaling
0
0 comments X

The pith

The optimal softmax temperature is uniquely determined by feature dimensionality.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper claims that the best temperature for the softmax function depends only on the number of dimensions in the feature representations. This would allow setting the temperature without any training or validation. The authors support this with a theoretical derivation and then introduce batch normalization plus fitted coefficients to create a practical empirical formula that works across models and datasets. They also add a correction term for the number of classes. If correct, this removes a common source of tuning effort in classification models.

Core claim

The optimal temperature T* is uniquely determined by the dimensionality of the feature representations, enabling training-free determination of T*. A set of temperature determination coefficients is optimized and a batch normalization layer is inserted before the output to stabilize the feature space, leading to an empirical formula that estimates T* and generalizes across tasks while improving performance.

What carries the argument

Analytical relationship linking optimal softmax temperature to feature dimensionality, stabilized by batch normalization and empirical coefficients.

If this is right

  • Classification accuracy increases without temperature tuning.
  • Temperature is set directly from feature dimensions using the formula.
  • The approach generalizes to different models, datasets, and complexities.
  • A corrective adjustment accounts for class count and task difficulty.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • This method may eliminate hyperparameter searches for temperature in many pipelines.
  • It suggests that other hyperparameters could be derived from architectural properties like dimension.
  • The stabilization might combine with other techniques for further gains.
  • Testing on out-of-distribution data would check robustness beyond the paper's experiments.

Load-bearing premise

The premise that batch normalization and the coefficients sufficiently stabilize the feature space for the formula to apply universally.

What would settle it

Compare the formula's suggested T* to the best T* found by search on a new model and dataset; large mismatch or no performance gain would disprove the generalizability.

Figures

Figures reproduced from arXiv: 2504.15594 by Shunsuke Sakai, Tatsuhito Hasegawa.

Figure 1
Figure 1. Figure 1: Sample outputs of the softmax function at various temperatures for a fixed model output [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Effect of temperature variation and label smoothing on cross-entropy loss. [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Outline of the flow of the general deep neural network models inserted a normalization layer. [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Change in accuracy relative to T in the CIFAR10 environment using VGG9 with M = 512. The results for T = 256 and T = 512 are below the drawing range. In [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Test accuracies [%] for each temperature parameter in various scenarios (without insertion of the normalization [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Test accuracies [%] for each temperature parameter in various scenarios (with BN insertion). Only CIFAR [PITH_FULL_IMAGE:figures/full_fig_p010_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Accuracy difference [%] between scenarios with and without BN insertion. Each value is calculated by [PITH_FULL_IMAGE:figures/full_fig_p010_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Estimation of test accuracy [%] for each temperature parameter (with BN insertion). The boxplots represent [PITH_FULL_IMAGE:figures/full_fig_p011_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Test accuracies [%] for each temperature parameter in different task difficulty scenarios using ResNet10. The [PITH_FULL_IMAGE:figures/full_fig_p013_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Test accuracies [%] for each temperature parameter in ISIC using ResNet50. [PITH_FULL_IMAGE:figures/full_fig_p017_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Distribution of the maximum softmax probability for varying class counts. [PITH_FULL_IMAGE:figures/full_fig_p019_11.png] view at source ↗
read the original abstract

In deep learning-based classification tasks, the softmax function's temperature parameter $T$ critically influences the output distribution and overall performance. This study presents a novel theoretical insight that the optimal temperature $T^*$ is uniquely determined by the dimensionality of the feature representations, thereby enabling training-free determination of $T^*$. Despite this theoretical grounding, empirical evidence reveals that $T^*$ fluctuates under practical conditions owing to variations in models, datasets, and other confounding factors. To address these influences, we propose and optimize a set of temperature determination coefficients that specify how $T^*$ should be adjusted based on the theoretical relationship to feature dimensionality. Additionally, we insert a batch normalization layer immediately before the output layer, effectively stabilizing the feature space. Building on these coefficients and a suite of large-scale experiments, we develop an empirical formula to estimate $T^*$ without additional training while also introducing a corrective scheme to refine $T^*$ based on the number of classes and task complexity. Our findings confirm that the derived temperature not only aligns with the proposed theoretical perspective but also generalizes effectively across diverse tasks, consistently enhancing classification performance and offering a practical, training-free solution for determining $T^*$.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript claims that the optimal softmax temperature T* is uniquely determined by the dimensionality of the feature representations, thereby enabling a training-free determination of T*. To address observed fluctuations with models, datasets, and task complexity, the authors introduce optimized temperature determination coefficients, insert a batch normalization layer immediately before the output layer, and develop an empirical formula from large-scale experiments that incorporates corrections based on the number of classes and task complexity.

Significance. If the central theoretical claim can be substantiated with a derivation, the work would provide a practical training-free method for setting softmax temperature that generalizes across models and domains, improving classification robustness. The large-scale empirical component offers some validation strength, but the current framing reduces the contribution to an empirically tuned adjustment rather than a dimensionally fixed analytical result.

major comments (2)
  1. [Abstract] Abstract: the claim that T* is 'uniquely determined by the dimensionality of the feature representations' is not supported by any derivation from the softmax function or loss landscape; the text immediately qualifies the uniqueness with the need for fitted coefficients and a BN layer, making the headline theoretical insight load-bearing yet unsubstantiated.
  2. [Abstract] Abstract: the temperature determination coefficients are optimized from the same large-scale experiments used to build the empirical formula, so the final T* estimate is a fitted adjustment rather than an independent derivation from feature dimensionality alone; this circularity directly affects the central claim of training-free, model- and domain-robust classification.
minor comments (2)
  1. [Abstract] Abstract: provide the explicit mathematical form of the empirical formula, including how it combines the dimensionality term with the class-count and task-complexity corrections.
  2. Clarify whether the BN layer is required for the theoretical relation to hold or only for the empirical generalization; this distinction is essential for assessing the method's scope.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive review. The comments highlight important points about the balance between our theoretical claim and the practical components of the work. We address each major comment below with clarifications drawn directly from the manuscript's analysis and experiments. We believe these responses, along with targeted revisions, will strengthen the presentation of the analytical relationship between feature dimensionality and optimal softmax temperature.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the claim that T* is 'uniquely determined by the dimensionality of the feature representations' is not supported by any derivation from the softmax function or loss landscape; the text immediately qualifies the uniqueness with the need for fitted coefficients and a BN layer, making the headline theoretical insight load-bearing yet unsubstantiated.

    Authors: We appreciate the referee's emphasis on substantiation. The manuscript derives the core relationship by analyzing the softmax output entropy and the scaling of feature norms in high-dimensional spaces, showing that T* scales proportionally with the square root of the feature dimensionality to maintain stable probability distributions. This analytical step is independent of later adjustments and is detailed in the theoretical section of the paper. The fitted coefficients and inserted batch normalization layer address observed deviations caused by model-specific factors, dataset variations, and task complexity, but they refine rather than replace the dimensional foundation. To address the concern about the abstract's framing, we will revise it to include a brief reference to the derivation basis while preserving the emphasis on the training-free aspect. revision: partial

  2. Referee: [Abstract] Abstract: the temperature determination coefficients are optimized from the same large-scale experiments used to build the empirical formula, so the final T* estimate is a fitted adjustment rather than an independent derivation from feature dimensionality alone; this circularity directly affects the central claim of training-free, model- and domain-robust classification.

    Authors: The coefficients are calibrated from the large-scale experiments to quantify how model, domain, and complexity factors modulate the base analytical scaling with dimensionality. This calibration does not create circularity for the central claim, as the primary dependence on feature dimensionality follows from the softmax analysis and holds as the starting point before any fitting. The experiments validate generalization and enable the corrective terms for number of classes and task complexity, supporting the training-free use across settings. We will revise the manuscript to more explicitly delineate the analytical derivation from the empirical calibration steps, thereby clarifying that the overall method remains training-free once the formula is established. revision: yes

Circularity Check

1 steps flagged

T* formula reduces to empirical fit of optimized coefficients on experimental data rather than pure dimensionality derivation

specific steps
  1. fitted input called prediction [Abstract]
    "This study presents a novel theoretical insight that the optimal temperature T* is uniquely determined by the dimensionality of the feature representations, thereby enabling training-free determination of T*. Despite this theoretical grounding, empirical evidence reveals that T* fluctuates under practical conditions owing to variations in models, datasets, and other confounding factors. To address these influences, we propose and optimize a set of temperature determination coefficients that specify how T* should be adjusted based on the theoretical relationship to feature dimensionality. ...we"

    The coefficients are explicitly optimized from large-scale experiments to handle fluctuations, after which an empirical formula is developed from those coefficients and the same experiments. The resulting T* estimate is therefore a direct output of the fitting process on the data rather than a derivation that holds from feature dimensionality by construction.

full rationale

The abstract asserts a novel theoretical insight that T* is uniquely determined by feature dimensionality, enabling training-free determination. However, it immediately notes fluctuations due to models/datasets and addresses them by optimizing temperature determination coefficients on large-scale experiments, then developing an empirical formula from those same coefficients and experiments. This structure means the practical T* estimate is constructed from the fitted adjustments rather than standing as an independent analytical result from dimensionality alone. The insertion of BN and corrective scheme for classes/task complexity further indicate the dimensionality-only claim requires empirical scaffolding to function, reducing the central prediction to a fitted input.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The central claim rests on an unproven theoretical uniqueness of T* from dimensionality plus a set of empirically optimized coefficients whose values are not supplied.

free parameters (1)
  • temperature determination coefficients
    A set of coefficients optimized to adjust the theoretical T* for practical variations in models and datasets.
axioms (1)
  • domain assumption Optimal temperature T* is uniquely determined by feature dimensionality
    Core theoretical premise stated without derivation details in the abstract.

pith-pipeline@v0.9.0 · 5891 in / 1096 out tokens · 122307 ms · 2026-05-22T18:35:46.715502+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

50 extracted references · 50 canonical work pages · 1 internal anchor

  1. [1]

    Krizhevsky, I

    A. Krizhevsky, I. Sutskever, and G. E. Hinton. Imagenet classification with deep convolutional neural networks. Advances in neural information processing systems, 25, 2012

    work page 2012

  2. [2]

    W. Liu, D. Anguelov, D. Erhan, C. Szegedy, S. Reed, C.-Y . Fu, and A. C. Berg. Ssd: Single shot multibox detectors. European conference on computer vision, pages 21–37, 2016

    work page 2016

  3. [3]

    Karras, S

    T. Karras, S. Laine, and T. Aila. A stylebased generator architecture for generative adversarial networks. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition, pages 4401–4410, 2019. 19 Running Title for Header

    work page 2019

  4. [4]

    Language models are few-shot learners

    Tom Brown, Benjamin Mann, Nick Ryder, Melanie Subbiah, Jared D Kaplan, Prafulla Dhariwal, Arvind Nee- lakantan, Pranav Shyam, Girish Sastry, Amanda Askell, Sandhini Agarwal, Ariel Herbert-V oss, Gretchen Krueger, Tom Henighan, Rewon Child, Aditya Ramesh, Daniel Ziegler, Jeffrey Wu, Clemens Winter, Chris Hesse, Mark Chen, Eric Sigler, Mateusz Litwin, Scott...

    work page 1901

  5. [5]

    Hinton, O

    G. Hinton, O. Vinyals, and J. Dean. Distilling the knowledge in a neural network. NIPS Deep Learning and Representation Learning Workshop (online), 2015

    work page 2015

  6. [6]

    C. Guo, G. Pleiss, Y . Sun, and K. Q. Weinberger. On calibration of modern neural networks.Proceedings of the 34th International Conference on Machine Learning, 70:1321–1330, 2017

    work page 2017

  7. [7]

    Deep learning for time series classification: a review

    Hassan Ismail Fawaz, Germain Forestier, Jonathan Weber, Lhassane Idoumghar, and Pierre-Alain Muller. Deep learning for time series classification: a review. Data Min. Knowl. Discov., 33(4):917–963, July 2019

    work page 2019

  8. [8]

    Temperature check: theory and practice for training models with softmax-cross-entropy losses

    Atish Agarwala, Samuel Stern Schoenholz, Jeffrey Pennington, and Yann Dauphin. Temperature check: theory and practice for training models with softmax-cross-entropy losses. Transactions on Machine Learning Research, 2023

    work page 2023

  9. [9]

    Optimal temperature parameter of softmax while training deep learning model in activity recognition

    Tatsuhito Hasegawa. Optimal temperature parameter of softmax while training deep learning model in activity recognition. Journal of Information Processing Society of Japan, 64(8):1182–1192, August 2023. (in Japanese)

    work page 2023

  10. [10]

    Exploring the impact of temperature scaling in softmax for classification and adversarial robustness, 2025

    Hao Xuan, Bokai Yang, and Xingyu Li. Exploring the impact of temperature scaling in softmax for classification and adversarial robustness, 2025

    work page 2025

  11. [11]

    Spectral metric for dataset complexity assessment

    Frédéric Branchaud-Charron, Andrew Achkar, and Pierre-Marc Jodoin. Spectral metric for dataset complexity assessment. In 2019 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 3210– 3219, 2019

    work page 2019

  12. [12]

    Ioffe and C

    S. Ioffe and C. Szegedy. Batch normalization: Accelerating deep network training by reducing internal covariate shift. ICML, 2015

    work page 2015

  13. [13]

    Meta knowledge distillation, 2022

    Jihao Liu, Boxiao Liu, Hongsheng Li, and Yu Liu. Meta knowledge distillation, 2022

    work page 2022

  14. [14]

    Curriculum temperature for knowledge distillation

    Zheng Li, Xiang Li, Lingfeng Yang, Borui Zhao, Renjie Song, Lei Luo, Jun Li, and Jian Yang. Curriculum temperature for knowledge distillation. In Proceedings of the Thirty-Seventh AAAI Conference on Artificial Intelligence, 2023

    work page 2023

  15. [15]

    Logit standardization in knowledge distillation

    Shangquan Sun, Wenqi Ren, Jingzhi Li, Rui Wang, and Xiaochun Cao. Logit standardization in knowledge distillation. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, pages 15731–15740, 2024

    work page 2024

  16. [16]

    Balanya, Juan Maroñas, and Daniel Ramos

    Sergio A. Balanya, Juan Maroñas, and Daniel Ramos. Adaptive temperature scaling for robust calibration of deep neural networks. Neural Computing and Applications, 36(14):8073–8095, May 2024

    work page 2024

  17. [17]

    Understanding the behaviour of contrastive loss

    Feng Wang and Huaping Liu. Understanding the behaviour of contrastive loss. In 2021 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 2495–2504, 2021

    work page 2021

  18. [18]

    Relaxed softmax: Efficient confidence auto-calibration for safe pedestrian detection

    Lukás Neumann, Andrew Zisserman, and A Vedaldi. Relaxed softmax: Efficient confidence auto-calibration for safe pedestrian detection. In Proceedings of the 2018 NIPS Workshop on Machine Learning for Intelligent Transportation Systems, 2018

    work page 2018

  19. [19]

    Heated-up softmax embedding, 2018

    Xu Zhang, Felix Xinnan Yu, Svebor Karaman, Wei Zhang, and Shih-Fu Chang. Heated-up softmax embedding, 2018

    work page 2018

  20. [20]

    Jimmy Lei Ba, Jamie Ryan Kiros, and Geoffrey E. Hinton. Layer normalization, 2016. arXiv:1607.06450

    work page internal anchor Pith review Pith/arXiv arXiv 2016

  21. [21]

    Gomez, Łukasz Kaiser, and Illia Polosukhin

    Ashish Vaswani, Noam Shazeer, Niki Parmar, Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. Attention is all you need. In Proceedings of the 31st International Conference on Neural Information Processing Systems, NIPS’17, page 6000–6010, 2017

    work page 2017

  22. [22]

    Szegedy, V

    C. Szegedy, V . Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna. Rethinking the inception architecture for computer vision. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 2818–2826, 2016

    work page 2016

  23. [23]

    When does label smoothing help? In Advances in Neural Information Processing Systems, volume 32, 2019

    Rafael Müller, Simon Kornblith, and Geoffrey E Hinton. When does label smoothing help? In Advances in Neural Information Processing Systems, volume 32, 2019

    work page 2019

  24. [24]

    Delving deep into label smoothing

    Chang-Bin Zhang, Peng-Tao Jiang, Qibin Hou, Yunchao Wei, Qi Han, Zhen Li, and Ming-Ming Cheng. Delving deep into label smoothing. Trans. Img. Proc., 30:5984–5996, jan 2021. 20 Running Title for Header

    work page 2021

  25. [25]

    Orr, and Klaus-Robert Müller

    Yann LeCun, Léon Bottou, Genevieve B. Orr, and Klaus-Robert Müller. Efficient backprop. In Neural Networks: Tricks of the Trade, This Book is an Outgrowth of a 1996 NIPS Workshop, page 9–50, Berlin, Heidelberg, 1998. Springer-Verlag

    work page 1996

  26. [26]

    Understanding the difficulty of training deep feedforward neural networks

    Xavier Glorot and Yoshua Bengio. Understanding the difficulty of training deep feedforward neural networks. In Proceedings of the Thirteenth International Conference on Artificial Intelligence and Statistics, volume 9, pages 249–256, 2010

    work page 2010

  27. [27]

    K. He, X. Zhang, S. Ren, and J. Sun. Delving deep into rectifiers: Surpassing human-level performance on imagenet classification. Proceedings of the IEEE International Conference on Computer Vision (ICCV), 2015

    work page 2015

  28. [28]

    Alex and H

    K. Alex and H. Geoffrey. Learning multiple layers of features from tiny images. Master’s thesis, University of Toronto, 2009

    work page 2009

  29. [29]

    An analysis of single-layer networks in unsupervised feature learning

    Adam Coates, Andrew Ng, and Honglak Lee. An analysis of single-layer networks in unsupervised feature learning. In Proceedings of the Fourteenth International Conference on Artificial Intelligence and Statistics , volume 15, pages 215–223, 2011

    work page 2011

  30. [30]

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei. Imagenet: A large-scale hierarchical image database. CVPR09, 2009

    work page 2009

  31. [31]

    Simonyan and A

    K. Simonyan and A. Zisserman. Very deep convolutional networks for large-scale image recognition.Proceedings of the International Conference on Learning Representations, pages 1–14, 2015

    work page 2015

  32. [32]

    Deep residual learning for image recognition

    Kaiming He, Xiangyu Zhang, Shaoqing Ren, and Jian Sun. Deep residual learning for image recognition. In 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), pages 770–778, 2016

    work page 2016

  33. [33]

    D. Han, J. Kim, and J.n Kim. Deep pyramidal residual networks. 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR, pages 6307–6315, 2017

    work page 2017

  34. [34]

    Loshchilov and F Hutter

    I. Loshchilov and F Hutter. Sgdr: Stochastic gradient descent with warm restarts. International Conference on Learning Representations, 2017

    work page 2017

  35. [35]

    Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces

    Rainer Storn and Kenneth Price. Differential evolution – a simple and efficient heuristic for global optimization over continuous spaces. Journal of Global Optimization, 11(4):341–359, Dec 1997

    work page 1997

  36. [36]

    EfficientNet: Rethinking model scaling for convolutional neural networks

    Mingxing Tan and Quoc Le. EfficientNet: Rethinking model scaling for convolutional neural networks. In Kamalika Chaudhuri and Ruslan Salakhutdinov, editors, Proceedings of the 36th International Conference on Machine Learning, volume 97 of Proceedings of Machine Learning Research, pages 6105–6114, 2019

    work page 2019

  37. [37]

    Designing network design spaces

    Ilija Radosavovic, Raj Prateek Kosaraju, Ross Girshick, Kaiming He, and Piotr Dollar. Designing network design spaces. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), June 2020

    work page 2020

  38. [38]

    A convnet for the 2020s

    Zhuang Liu, Hanzi Mao, Chao-Yuan Wu, Christoph Feichtenhofer, Trevor Darrell, and Saining Xie. A convnet for the 2020s. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 11976–11986, June 2022

    work page 2022

  39. [39]

    An image is worth 16x16 words: Transformers for image recognition at scale

    Alexey Dosovitskiy, Lucas Beyer, Alexander Kolesnikov, Dirk Weissenborn, Xiaohua Zhai, Thomas Unterthiner, Mostafa Dehghani, Matthias Minderer, Georg Heigold, Sylvain Gelly, Jakob Uszkoreit, and Neil Houlsby. An image is worth 16x16 words: Transformers for image recognition at scale. In International Conference on Learning Representations, 2021

    work page 2021

  40. [40]

    Z. Liu, Y . Lin, Y . Cao, H. Hu, Y . Wei, Z. Zhang, S. Lin, and B. Guo. Swin transformer: Hierarchical vision transformer using shifted windows. In 2021 IEEE/CVF International Conference on Computer Vision (ICCV), pages 9992–10002, 2021

    work page 2021

  41. [41]

    Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories

    Li Fei-Fei, Rob Fergus, and Pietro Perona. Learning generative visual models from few training examples: An incremental bayesian approach tested on 101 object categories. Computer vision and Image understanding , 106(1):59–70, 2007

    work page 2007

  42. [42]

    Detection of traffic signs in real-world images: The German Traffic Sign Detection Benchmark

    Sebastian Houben, Johannes Stallkamp, Jan Salmen, Marc Schlipsing, and Christian Igel. Detection of traffic signs in real-world images: The German Traffic Sign Detection Benchmark. In International Joint Conference on Neural Networks, number 1288, 2013

    work page 2013

  43. [43]

    Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification

    Patrick Helber, Benjamin Bischke, Andreas Dengel, and Damian Borth. Eurosat: A novel dataset and deep learning benchmark for land use and land cover classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, 2019

    work page 2019

  44. [44]

    The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions

    Philipp Tschandl, Cliff Rosendahl, and Harald Kittler. The ham10000 dataset, a large collection of multi-source dermatoscopic images of common pigmented skin lesions. Scientific Data, 5(1):180161, Aug 2018. 21 Running Title for Header

    work page 2018

  45. [45]

    Noel C. F. Codella, David Gutman, M. Emre Celebi, Brian Helba, Michael A. Marchetti, Stephen W. Dusza, Aadi Kalloo, Konstantinos Liopyris, Nabin Mishra, Harald Kittler, and Allan Halpern. Skin lesion analysis toward melanoma detection: A challenge at the 2017 international symposium on biomedical imaging (isbi), hosted by the international skin imaging co...

    work page 2017

  46. [46]

    Carlos Hernández-Pérez, Marc Combalia, Sebastian Podlipnik, Noel C. F. Codella, Veronica Rotemberg, Allan C. Halpern, Ofer Reiter, Cristina Carrera, Alicia Barreiro, Brian Helba, Susana Puig, Veronica Vilaplana, and Josep Malvehy. Bcn20000: Dermoscopic lesions in the wild. Scientific Data, 11(1):641, Jun 2024

    work page 2024

  47. [47]

    Nilsback and A

    M-E. Nilsback and A. Zisserman. Automated flower classification over a large number of classes. In Proceedings of the Indian Conference on Computer Vision, Graphics and Image Processing, Dec 2008

    work page 2008

  48. [48]

    Omkar M Parkhi, Andrea Vedaldi, Andrew Zisserman, and C. V . Jawahar. Cats and dogs. In2012 IEEE Conference on Computer Vision and Pattern Recognition, pages 3498–3505, 2012

    work page 2012

  49. [49]

    Describing textures in the wild

    Mircea Cimpoi, Subhransu Maji, Iasonas Kokkinos, Sammy Mohamed, and Andrea Vedaldi. Describing textures in the wild. In Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2014

    work page 2014

  50. [50]

    C. Wah, S. Branson, P. Welinder, P. Perona, and S. Belongie. The caltech-ucsd birds-200-2011 dataset. Technical Report CNS-TR-2011-001, California Institute of Technology, 2011. 22

    work page 2011