Deconvolutional networks

By Matthew D. Zeiler, Dilip Krishnan, Graham W. Taylor, Rob Fergus. 2010

1,689 citations recorded by the source corpus.

Building robust low and mid-level image representations, beyond edge primitives, is a long-standing goal in vision. Many existing feature detectors spatially pool edge information which destroys cues such as edge intersections, parallelism and symmetry. We present a learning framework where features that capture these mid-level cues spontaneously emerge from image data. Our approach is based on the convolutional decomposition of images under a spar-sity constraint and is totally unsupervised. By building a hierarchy of such decompositions we can learn rich feature sets that are a robust image representation for both the analysis and synthesis of images.

Subfields: Computer Vision and Pattern Recognition, Media Technology, Generative Models

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computer Vision and Pattern Recognition

In Media Technology

In Generative Models

Citations in the corpus

Beyond Skip Connections: Top-Down Modulation for Object Detection (2016, methodology): “Deconvolutional [36, 55] and encoder-decoder networks [3] are often used to generate such finer features by using the ‘unpool’ operation.”
Learning to Refine Object Segments (2016, methodology): “Deconvolutional networks: [40] proposed to invert the pooling process in a CNN to generate progressively higher resolution input images by storing the ‘switch’ variables from the pooling operation.”
Towards Conceptual Compression (2016, background): “…auto-encoders and sparse coding (Kavukcuoglu et al., 2010; Le, 2013), denoising auto-encoders (Vincent et al., 2010), deconvolutional networks (Zeiler et al., 2010), restricted Boltzmann machines (Hin-ton & Salakhutdinov, 2006), deep Boltzmann machines (Salakhutdinov & Hinton, 2009),…”
Pedestrian Detection with Unsupervised Multi-stage Feature Learning (2012, methodology): “[20, 39] have introduced convolutional sparse modeling formulations for feature learning and object recognition and we use the Convolutional Predictive Sparse Decomposition (CPSD) model proposed in [20] since it is the only convolutional sparse codingmodel providing a fast predictor function that is”
Generating Videos with Scene Dynamics (2016, methodology): “recognizing actions in video with deep networks, but apply them for video generation instead. We use spatio-temporal 3D convolutions to model videos [40], but we use fractionally strided convolutions [52] instead because we are interested in generation. We also use two-streams to model video [34], b”
Embed to Control: A Locally Linear Latent Dynamics Model for Control from Raw Images (2015, methodology): “We consider two different network types for our model: Standard fully connected neural networks with up to three layers which work well for moderately sized images and are used for the planar and swing-up experiments; A deep convolutional network for the encoder in combination with an “up-convolutio”
Representation Learning: A Review and New Perspectives (2012, methodology): “Other unsupervised feature learning approaches that were adapted to the convolutional setting include PSD [110], [111], [106], [84], a convolutional version of sparse coding called deconvolutional networks [226], topographic ICA [122], and mPoT, which Kivinen and Williams [113] applied to modeling n”
Stacked What-Where Auto-encoders (2015, methodology): “An instantiation of SWWAE is essentially a convolutional net (Convnet) [14] coupled with a deconvolutional net (Deconvnet) [30].”
SegNet: A Deep Convolutional Encoder-Decoder Architecture for Image Segmentation (2015, methodology): “A somewhat similar decoding technique is used for visualizing trained convolutional networks [47] for classification.”
Deep Convolutional Inverse Graphics Network (2015, methodology): “Another piece of related work is the deconvolution network by Zeiler et al. [32] which is used to produce image representations using convolutional decomposition of images under a sparsity constraint.”
Stacked Hourglass Networks for Human Pose Estimation (2016, methodology): “eated bottom-up, topdown inference by stacking two hourglasses. Our hourglass module before stacking is also related to the convolutiondeconvolution architecture [39,40,41], which deploys a DeconvNet [42] to do pixel-wise prediction. Noh et al. [39] used the conv-deconv architecture to do semantic s”
Explainable Artificial Intelligence (XAI): Concepts, Taxonomies, Opportunities and Challenges toward Responsible AI (2019, positive): “The authors in [291] used Deconvnet, a network designed previously by the same authors [142] that, when fed with a feature map from a selected layer, reconstructs the maximum activations.”
Explanation Methods in Deep Learning: Users, Values, Concerns and Challenges (2018, neutral): “For example, deconvolutional networks [57] can be applied to models that make use of convolutions to extract features from input images.”
What do we need to build explainable AI systems for the medical domain? (2017, methodology): “One technique, which is highly interesting for the medical domain, e.g. for images generated by digital pathology (which are orders of magnitudes larger than e.g. radiological images) (Holzinger et al., 2017b) is the use of deconvolutional networks (Zeiler et al., 2010).”
Hierarchical Surface Prediction for 3D Object Reconstruction (2017, methodology): “An up-convolutional also called de-convolutional decoder architecture [10], [39] predicts the voxel block octree.”