--- title: "Visualizing and Understanding Convolutional — Zeiler 2013" description: "Large convnets such as the Krizhevsky et al. (2012) ImageNet model classify images very well, but it is hard to see what their inner layers compute.…" authors: ["Matthew D Zeiler", "Rob Fergus"] year: 2013 venue: "arXiv:1311.2901" source: "https://arxiv.org/abs/1311.2901" doi: "10.48550/arxiv.1311.2901" arxiv: "1311.2901" licence: "arXiv — non-exclusive licence to distribute" licence_url: "http://arxiv.org/licenses/nonexclusive-distrib/1.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/zeiler-2013-visualizing-understanding-convolutional-networks/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zeiler-2013-visualizing-understanding-convolutional-networks/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2013](https://md.research.elamur.ai/free/en/papers/2013/page/2/) / Visualizing and Understanding Convolutional Networks # Visualizing and Understanding Convolutional Networks **Authors:** [Matthew D Zeiler](https://md.research.elamur.ai/free/en/authors/matthew-d-zeiler/), [Rob Fergus](https://md.research.elamur.ai/free/en/authors/rob-fergus/) · **Year:** 2013 · **Venue:** arXiv:1311.2901 · **Licence:** [arXiv — non-exclusive licence to distribute](http://arxiv.org/licenses/nonexclusive-distrib/1.0/) **Links:** [Publisher's page](https://arxiv.org/abs/1311.2901) · [PDF](https://arxiv.org/pdf/1311.2901) · [DOI 10.48550/arxiv.1311.2901](https://doi.org/10.48550/arxiv.1311.2901) · [arXiv 1311.2901](https://arxiv.org/abs/1311.2901) · [OpenAlex](https://openalex.org/W2952186574) **Citations:** 394 citations (openalex-snapshot · 2026-10-14) · [255](#cited-by) in our library, of the 140673 papers whose bibliographies we read ## Summary Large convnets such as the Krizhevsky et al. (2012) ImageNet model classify images very well, but it is hard to see what their inner layers compute. Without that view, improving them comes down to trial and error. Zeiler and Fergus propose a diagnostic tool. A Deconvolutional Network (deconvnet) is attached to a trained convnet and maps any chosen activation back to pixel space. It does this by unpooling with recorded max-location "switches", rectifying, and applying the transposed filters. They add occlusion-based sensitivity tests and an ablation study of individual layers. Their model was trained on ImageNet 2012 (1.3M images, 1000 classes). The visualizations show that the features become more compositional, more invariant and more class-specific in higher layers. They also exposed defects in the first two layers, and fixing these gave an architecture that beats Krizhevsky et al. on ImageNet. When only the softmax is retrained, the features transfer well and set new records on Caltech-101 and Caltech-256. This work helped establish supervised ImageNet pre-training as a general-purpose feature extractor and made convnet interpretability a practical debugging method. ## Key points - The deconvnet is not trained. It acts only as a probe that reuses the convnet's own filters (flipped) and the pooling switches recorded for a given input. - Training setup: SGD with batch 128, learning rate 10⁻², momentum 0.9, dropout 0.5 in layers 6–7, and filter RMS renormalization to radius 10⁻¹. Training ran 70 epochs, about 12 days on one GTX580. - Caltech-101 (30 images/class): 86.5% accuracy against the previous best of 84.3%. The same network trained from scratch reaches only 46.5%. - Caltech-256 (60 images/class): 74.2% against 55.2% for Bo et al. (2013). About 6 training images per class are enough to beat the previous best method, which used roughly ten times as many. - PASCAL 2012: mean accuracy is 79.0% against 82.2% for the leading method, but the model wins on 5 classes. Layer-wise SVM/softmax probes show accuracy rising steadily with depth, and the ablation shows that overall network depth matters more than any single layer. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Other](https://md.research.elamur.ai/free/en/shelves/other/) ## Cited by Cited by 255 papers of our library cite it (of the 140673 whose bibliographies we read); 24 of them have a page here, 231 not yet. - [A Diagnostic Study of Explainability Techniques for Text Classification](https://md.research.elamur.ai/free/en/papers/atanasova-2020-diagnostic-study-explainability-techniques-text/) - [Interpreting Neural Networks to Improve Politeness Comprehension](https://md.research.elamur.ai/free/en/papers/aubakirova-2016-interpreting-neural-networks-improve-politeness/) - [On the Opportunities and Risks of Foundation Models](https://md.research.elamur.ai/free/en/papers/bommasani-2021-opportunities-risks-foundation-models/) - [Visualizing and Measuring the Geometry of BERT](https://md.research.elamur.ai/free/en/papers/coenen-2019-visualizing-measuring-geometry-bert/) - [Very Deep Convolutional Networks for Text Classification](https://md.research.elamur.ai/free/en/papers/conneau-2017-very-deep-convolutional-networks-text/) - [Long-term Recurrent Convolutional Networks for Visual Recognition and Description](https://md.research.elamur.ai/free/en/papers/donahue-2014-long-term-recurrent-convolutional-networks/) - [Visualizing and Understanding Neural Models in NLP](https://md.research.elamur.ai/free/en/papers/li-2016-visualizing-understanding-neural-models-nlp/) - [ILDC for CJPE: Indian Legal Documents Corpus for Court Judgment Prediction and Explanation](https://md.research.elamur.ai/free/en/papers/malik-2021-ildc-cjpe-indian-legal-documents/) - [Dissecting Contextual Word Embeddings: Architecture and Representation](https://md.research.elamur.ai/free/en/papers/peters-2018-dissecting-contextual-word-embeddings-architecture/) - [Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks](https://md.research.elamur.ai/free/en/papers/radford-2015-unsupervised-representation-learning-deep-convolutional/) - [Learning to Generate Reviews and Discovering Sentiment](https://md.research.elamur.ai/free/en/papers/radford-2017-learning-generate-reviews-discovering-sentiment/) - [Learning Transferable Visual Models From Natural Language Supervision](https://md.research.elamur.ai/free/en/papers/radford-2021-learning-transferable-visual-models-natural/) - [DeepSpeed-MoE: Advancing Mixture-of-Experts Inference and Training to Power Next-Generation AI Scale](https://md.research.elamur.ai/free/en/papers/rajbhandari-2022-deepspeed-moe-advancing-mixture-experts/) - [FaceNet: A Unified Embedding for Face Recognition and Clustering](https://md.research.elamur.ai/free/en/papers/schroff-2015-facenet-unified-embedding-face-recognition/) - [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization](https://md.research.elamur.ai/free/en/papers/selvaraju-2016-grad-cam-visual-explanations-deep/) - [Learning Important Features Through Propagating Activation Differences](https://md.research.elamur.ai/free/en/papers/shrikumar-2017-learning-important-features-through-propagating/) - [Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps](https://md.research.elamur.ai/free/en/papers/simonyan-2013-deep-inside-convolutional-networks-visualising/) - [Very Deep Convolutional Networks for Large-Scale Image Recognition](https://md.research.elamur.ai/free/en/papers/simonyan-2014-very-deep-convolutional-networks-large/) - [Intriguing properties of neural networks](https://md.research.elamur.ai/free/en/papers/szegedy-2013-intriguing-properties-neural-networks/) - [深度知识追踪模型综述和性能比较](https://md.research.elamur.ai/free/en/papers/wang-2022-performance-comparison-deep/) - [Inference is Everything: Recasting Semantic Resources into a Unified Evaluation Framework](https://md.research.elamur.ai/free/en/papers/white-2017-inference-is-everything-recasting-semantic/) - [DeeBERT: Dynamic Early Exiting for Accelerating BERT Inference](https://md.research.elamur.ai/free/en/papers/xin-2020-deebert-dynamic-early-exiting-accelerating/) - [Socratic Models: Composing Zero-Shot Multimodal Reasoning with Language](https://md.research.elamur.ai/free/en/papers/zeng-2022-socratic-models-composing-zero-shot/) - [Representation Engineering: A Top-Down Approach to AI Transparency](https://md.research.elamur.ai/free/en/papers/zou-2023-representation-engineering-top-down-approach/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @misc{zeiler2013visualizing, title = {{Visualizing and Understanding Convolutional Networks}}, author = {Zeiler, Matthew D and Fergus, Rob}, year = {2013}, doi = {10.48550/arxiv.1311.2901}, eprint = {1311.2901}, archivePrefix = {arXiv}, url = {https://doi.org/10.48550/arxiv.1311.2901} } ``` - [← previous Veridicity annotation in the lexicon? A look at factive adjectives](https://md.research.elamur.ai/free/en/papers/zaenen-2013-veridicity-annotation-lexicon-look-factive/) - [next → Weakly Supervised Learning of Semantic Parsers for Mapping Instructions to Actions](https://md.research.elamur.ai/free/en/papers/artzi-2013-weakly-supervised-learning-semantic-parsers/) - [↑ back to papers of 2013](https://md.research.elamur.ai/free/en/papers/2013/page/2/) ## References 23 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Bengio, Y., Lamblin, P., Popovici, D., and Larochelle, H. Greedy layer-wise training of deep networks. In NIPS, pp. 153–160, 2007. [find](https://scholar.google.com/scholar?q=Bengio%2C+Y%2E%2C+Lamblin%2C+P%2E%2C+Popovici%2C+D%2E%2C+and+Larochelle%2C+H%2E+Greedy+layer%2Dwise+training+of+deep+networks%2E+In+NIPS%2C+pp%2E+153%E2%80%93160%2C+2007%2E) 2. Berkes, P. and Wiskott, L. On the analysis and interpretation of inhomogeneous quadratic forms as receptive fields. Neural Computation, 2006. [find](https://scholar.google.com/scholar?q=Berkes%2C+P%2E+and+Wiskott%2C+L%2E+On+the+analysis+and+interpretation+of+inhomogeneous+quadratic+forms+as+receptive+fields%2E+Neural+Computation%2C+2006%2E) 3. Bo, L., Ren, X., and Fox, D. Multipath sparse coding using hierarchical matching pursuit. In CVPR, 2013. [find](https://scholar.google.com/scholar?q=Bo%2C+L%2E%2C+Ren%2C+X%2E%2C+and+Fox%2C+D%2E+Multipath+sparse+coding+using+hierarchical+matching+pursuit%2E+In+CVPR%2C+2013%2E) 4. Ciresan, D. C., Meier, J., and Schmidhuber, J. Multi-column deep neural networks for image classification. In CVPR, 2012. [find](https://scholar.google.com/scholar?q=Ciresan%2C+D%2E+C%2E%2C+Meier%2C+J%2E%2C+and+Schmidhuber%2C+J%2E+Multi%2Dcolumn+deep+neural+networks+for+image+classification%2E+In+CVPR%2C+2012%2E) 5. Dalal, N. and Triggs, B. Histograms of oriented gradients for pedestrian detection. In CVPR, 2005. [find](https://scholar.google.com/scholar?q=Dalal%2C+N%2E+and+Triggs%2C+B%2E+Histograms+of+oriented+gradients+for+pedestrian+detection%2E+In+CVPR%2C+2005%2E) 6. Donahue, J., Jia, Y., Vinyals, O., Hoffman, J., Zhang, N., Tzeng, E., and Darrell, T. DeCAF: A deep convolutional activation feature for generic visual recognition. In arXiv:1310.1531, 2013. [find](https://scholar.google.com/scholar?q=Donahue%2C+J%2E%2C+Jia%2C+Y%2E%2C+Vinyals%2C+O%2E%2C+Hoffman%2C+J%2E%2C+Zhang%2C+N%2E%2C+Tzeng%2C+E%2E%2C+and+Darrell%2C+T%2E+DeCAF%3A+A+deep+convolutional+activation+feature+for+generic+visual+recognition%2E+In+arXiv%3A1310%2E1531%2C+2013%2E) 7. Erhan, D., Bengio, Y., Courville, A., and Vincent, P. Visualizing higher-layer features of a deep network. In Technical report, University of Montreal, 2009. [find](https://scholar.google.com/scholar?q=Erhan%2C+D%2E%2C+Bengio%2C+Y%2E%2C+Courville%2C+A%2E%2C+and+Vincent%2C+P%2E+Visualizing+higher%2Dlayer+features+of+a+deep+network%2E+In+Technical+report%2C+University+of+Montreal%2C+2009%2E) 8. Fei-fei, L., Fergus, R., and Perona, P. One-shot learning of object categories. IEEE Trans. PAMI, 2006. [find](https://scholar.google.com/scholar?q=Fei%2Dfei%2C+L%2E%2C+Fergus%2C+R%2E%2C+and+Perona%2C+P%2E+One%2Dshot+learning+of+object+categories%2E+IEEE+Trans%2E+PAMI%2C+2006%2E) 9. Griffin, G., Holub, A., and Perona, P. The caltech 256. In Caltech Technical Report, 2006. [find](https://scholar.google.com/scholar?q=Griffin%2C+G%2E%2C+Holub%2C+A%2E%2C+and+Perona%2C+P%2E+The+caltech+256%2E+In+Caltech+Technical+Report%2C+2006%2E) 10. Gunji, N., Higuchi, T., Yasumoto, K., Muraoka, H., Ushiku, Y., Harada, T., and Kuniyoshi, Y. Classification entry. In Imagenet Competition, 2012. [find](https://scholar.google.com/scholar?q=Gunji%2C+N%2E%2C+Higuchi%2C+T%2E%2C+Yasumoto%2C+K%2E%2C+Muraoka%2C+H%2E%2C+Ushiku%2C+Y%2E%2C+Harada%2C+T%2E%2C+and+Kuniyoshi%2C+Y%2E+Classification+entry%2E+In+Imagenet+Competition%2C+2012%2E) 11. Hinton, G. E., Osindero, S., and The, Y. A fast learning algorithm for deep belief nets. Neural Computation, 18:1527–1554, 2006. [find](https://scholar.google.com/scholar?q=Hinton%2C+G%2E+E%2E%2C+Osindero%2C+S%2E%2C+and+The%2C+Y%2E+A+fast+learning+algorithm+for+deep+belief+nets%2E+Neural+Computation%2C+18%3A1527%E2%80%931554%2C+2006%2E) 12. Hinton, G.E., Srivastave, N., Krizhevsky, A., Sutskever, I., and Salakhutdinov, R. R. Improving neural networks by preventing co-adaptation of feature detectors. arXiv:1207.0580, 2012. [find](https://scholar.google.com/scholar?q=Hinton%2C+G%2EE%2E%2C+Srivastave%2C+N%2E%2C+Krizhevsky%2C+A%2E%2C+Sutskever%2C+I%2E%2C+and+Salakhutdinov%2C+R%2E+R%2E+Improving+neural+networks+by+preventing+co%2Dadaptation+of+feature+detectors%2E+arXiv%3A1207%2E0580%2C+2012%2E) 13. Jarrett, K., Kavukcuoglu, K., Ranzato, M., and Le-Cun, Y. What is the best multi-stage architecture for object recognition? In ICCV, 2009. [find](https://scholar.google.com/scholar?q=Jarrett%2C+K%2E%2C+Kavukcuoglu%2C+K%2E%2C+Ranzato%2C+M%2E%2C+and+Le%2DCun%2C+Y%2E+What+is+the+best+multi%2Dstage+architecture+for+object+recognition%3F+In+ICCV%2C+2009%2E) 14. Jianchao, Y., Kai, Y., Yihong, G., and Thomas, H. Linear spatial pyramid matching using sparse coding for image classification. In CVPR, 2009. [find](https://scholar.google.com/scholar?q=Jianchao%2C+Y%2E%2C+Kai%2C+Y%2E%2C+Yihong%2C+G%2E%2C+and+Thomas%2C+H%2E+Linear+spatial+pyramid+matching+using+sparse+coding+for+image+classification%2E+In+CVPR%2C+2009%2E) 15. Krizhevsky, A., Sutskever, I., and Hinton, G.E. Imagenet classification with deep convolutional neural networks. In NIPS, 2012. [find](https://scholar.google.com/scholar?q=Krizhevsky%2C+A%2E%2C+Sutskever%2C+I%2E%2C+and+Hinton%2C+G%2EE%2E+Imagenet+classification+with+deep+convolutional+neural+networks%2E+In+NIPS%2C+2012%2E) 16. Le, Q. V., Ngiam, J., Chen, Z., Chia, D., Koh, P., and Ng, A. Y. Tiled convolutional neural networks. In NIPS, 2010. [find](https://scholar.google.com/scholar?q=Le%2C+Q%2E+V%2E%2C+Ngiam%2C+J%2E%2C+Chen%2C+Z%2E%2C+Chia%2C+D%2E%2C+Koh%2C+P%2E%2C+and+Ng%2C+A%2E+Y%2E+Tiled+convolutional+neural+networks%2E+In+NIPS%2C+2010%2E) 17. LeCun, Y., Boser, B., Denker, J. S., Henderson, D., Howard, R. E., Hubbard, W., and Jackel, L. D. Backpropagation applied to handwritten zip code recognition. Neural Comput., 1(4):541–551, 1989. [find](https://scholar.google.com/scholar?q=LeCun%2C+Y%2E%2C+Boser%2C+B%2E%2C+Denker%2C+J%2E+S%2E%2C+Henderson%2C+D%2E%2C+Howard%2C+R%2E+E%2E%2C+Hubbard%2C+W%2E%2C+and+Jackel%2C+L%2E+D%2E+Backpropagation+applied+to+handwritten+zip+code+recognition%2E+Neural+Comput%2E%2C+1%284%29%3A541%E2%80%93551%2C+1989%2E) 18. Sande, K., Uijlings, J., Snoek, C., and Smeulders, A. Hybrid coding for selective search. In PASCAL VOC Classification Challenge 2012, 2012. [find](https://scholar.google.com/scholar?q=Sande%2C+K%2E%2C+Uijlings%2C+J%2E%2C+Snoek%2C+C%2E%2C+and+Smeulders%2C+A%2E+Hybrid+coding+for+selective+search%2E+In+PASCAL+VOC+Classification+Challenge+2012%2C+2012%2E) 19. Sohn, K., Jung, D., Lee, H., and Hero III, A. Efficient learning of sparse, distributed, convolutional feature representations for object recognition. In ICCV, 2011. [find](https://scholar.google.com/scholar?q=Sohn%2C+K%2E%2C+Jung%2C+D%2E%2C+Lee%2C+H%2E%2C+and+Hero+III%2C+A%2E+Efficient+learning+of+sparse%2C+distributed%2C+convolutional+feature+representations+for+object+recognition%2E+In+ICCV%2C+2011%2E) 20. Torralba, A. and Efros, A. A. Unbiased look at dataset bias. In CVPR, 2011. [find](https://scholar.google.com/scholar?q=Torralba%2C+A%2E+and+Efros%2C+A%2E+A%2E+Unbiased+look+at+dataset+bias%2E+In+CVPR%2C+2011%2E) 21. Vincent, P., Larochelle, H., Bengio, Y., and Manzagol, P. A. Extracting and composing robust features with denoising autoencoders. In ICML, pp. 1096– 1103, 2008. [find](https://scholar.google.com/scholar?q=Vincent%2C+P%2E%2C+Larochelle%2C+H%2E%2C+Bengio%2C+Y%2E%2C+and+Manzagol%2C+P%2E+A%2E+Extracting+and+composing+robust+features+with+denoising+autoencoders%2E+In+ICML%2C+pp%2E+1096%E2%80%93+1103%2C+2008%2E) 22. Yan, S., Dong, J., Chen, Q., Song, Z., Pan, Y., Xia, W., Huang, Z., Hua, Y., and Shen, S. Generalized hierarchical matching for sub-category aware object classification. In PASCAL VOC Classification Challenge 2012, 2012. [find](https://scholar.google.com/scholar?q=Yan%2C+S%2E%2C+Dong%2C+J%2E%2C+Chen%2C+Q%2E%2C+Song%2C+Z%2E%2C+Pan%2C+Y%2E%2C+Xia%2C+W%2E%2C+Huang%2C+Z%2E%2C+Hua%2C+Y%2E%2C+and+Shen%2C+S%2E+Generalized+hierarchical+matching+for+sub%2Dcategory+aware+object+classification%2E+In+PASCAL+VOC+Classification+C) 23. Zeiler, M., Taylor, G., and Fergus, R. Adaptive deconvolutional networks for mid and high level feature learning. In ICCV, 2011. [find](https://scholar.google.com/scholar?q=Zeiler%2C+M%2E%2C+Taylor%2C+G%2E%2C+and+Fergus%2C+R%2E+Adaptive+deconvolutional+networks+for+mid+and+high+level+feature+learning%2E+In+ICCV%2C+2011%2E)