Filters








73 Hits in 4.2 sec

Amortised MAP Inference for Image Super-resolution [article]

Casper Kaae Sønderby, Jose Caballero, Lucas Theis, Wenzhe Shi, Ferenc Huszár
2017 arXiv   pre-print
Here we introduce new methods for amortised MAP inference whereby we calculate the MAP estimate directly using a convolutional neural network.  ...  Image super-resolution (SR) is an underdetermined inverse problem, where a large number of plausible high-resolution images can explain the same downsampled image.  ...  DENSITY GUIDED SUPER-RESOLUTION As a more direct baseline model for amortised MAP inference we fit a tractable, yet powerful density model to p Y using maximum likelihood, and then use cross entropy with  ... 
arXiv:1610.04490v3 fatcat:fv4thg353ncx5glwxju6ohg6re

Amortised MAP Inference for Image Super-Resolution [article]

Ferenc Huszar, Casper Kaae Sønderby, Jose Caballer, Lucas Theis, Wenzhe Shi, Apollo-University Of Cambridge Repository, Apollo-University Of Cambridge Repository
2020
Here we introduce new methods for amortised MAP inference whereby we calculate the MAP estimate directly using a convolutional neural network.  ...  Image super-resolution (SR) is an underdetermined inverse problem, where a large number of plausible high resolution images can explain the same downsampled image.  ...  DENSITY GUIDED SUPER-RESOLUTION As a more direct baseline model for amortised MAP inference we fit a tractable, yet powerful density model to p Y using maximum likelihood, and then use cross entropy with  ... 
doi:10.17863/cam.51994 fatcat:zpb7ebqrczbi3cdzbktn2ix5am

Recurrent Generative Adversarial Networks for Proximal Learning and Automated Compressive Image Recovery [article]

Morteza Mardani, Hatef Monajemi, Vardan Papyan, Shreyas Vasanawala, David Donoho, John Pauly
2017 arXiv   pre-print
For image superresolution, our preliminary results indicate that modeling the denoising proximal demands deep ResNets.  ...  Recovering images from undersampled linear measurements typically leads to an ill-posed linear inverse problem, that asks for proper statistical priors.  ...  Single image super-resolution More evaluations are performed for super-resolving natural images.  ... 
arXiv:1711.10046v1 fatcat:rpx3hzkl6ffpndbzzkmv5einnu

How to Reach Real-Time AI on Consumer Devices? Solutions for Programmable and Custom Architectures [article]

Stylianos I. Venieris and Ioannis Panopoulos and Ilias Leontiadis and Iakovos S. Venieris
2021 arXiv   pre-print
Our findings provide illustrative examples of AI systems that do not overburden mobile hardware, while also indicating how they can improve inference accuracy.  ...  The unprecedented performance of deep neural networks (DNNs) has led to large strides in various Artificial Intelligence (AI) inference tasks, such as object and speech recognition.  ...  Such a NAS-generated model is TPSR [44] , a compact DNN for the task of image super-resolution.  ... 
arXiv:2106.15021v1 fatcat:b25jifosajeuba57qxiaockmg4

Variational Inference using Implicit Distributions [article]

Ferenc Huszár
2017 arXiv   pre-print
These models are highly expressive and we argue they can prove just as useful for variational inference (VI) as they are for generative modelling.  ...  This paper provides a unifying review of existing algorithms establishing connections between variational autoencoders, adversarially learned inference, operator VI, GAN-based image reconstruction, and  ...  Sønderby et al. (2017) used this insight to build a denoiser-based inference algorithm for image super-resolution and connected it to amortised maximum a posteriori (MAP) inference.  ... 
arXiv:1702.08235v1 fatcat:ytvca7oakfa2rinnvczkyrlewm

Viscos Flows: Variational Schur Conditional Sampling With Normalizing Flows [article]

Vincent Moens, Aivar Sootla, Haitham Bou Ammar, Jun Wang
2021 arXiv   pre-print
Our numerical results indicate that our sampling method can be successfully applied to invertible residual networks for inference and classification.  ...  We present a method for conditional sampling for pre-trained normalizing flows when only part of an observation is available.  ...  This is met in a wide range of applications, from few shots learning (Wang et al., 2020) to super resolution image generation (Bashir et al., 2021) , high-dimensional Bayesian optimisation (Moriconi  ... 
arXiv:2107.02474v3 fatcat:f4km4nz3a5dwjgajwh4iyrtuo4

High-resolution Deep Convolutional Generative Adversarial Networks [article]

J. D. Curtó and I. C. Zarza and Fernando de la Torre and Irwin King and Michael R. Lyu
2020 arXiv   pre-print
, HDCGAN, that incorporates current state-of-the-art techniques for this effect.  ...  A novel bias-free dataset, Curtó Zarza, containing human faces from different ethnical groups in a wide variety of illumination conditions and image resolutions is introduced.  ...  Amortised map by a telescope ζ while keeping all convolutional filters unchanged, inference for image super-resolution. ICLR (2017).  ... 
arXiv:1711.06491v18 fatcat:yfvv3mge3vgbrffy3gts3ae45a

Interpreting Spatially Infinite Generative Models [article]

Chaochao Lu, Richard E. Turner, Yingzhen Li, Nate Kushman
2020 arXiv   pre-print
resolution training images.  ...  Experiments on world map generation, panoramic images and texture synthesis verify the ability of ∞-GAN to efficiently generate images of arbitrary size.  ...  Amortised map inference for image superresolution. arXiv preprint arXiv:1610.04490, 2016. Ulyanov, D., Lebedev, V., Lempitsky, V., et al.  ... 
arXiv:2007.12411v1 fatcat:bafu5oyxtfakxbi6lkaerxbwzy

Conditional Generative Adversarial Networks for Domain Transfer: A Survey

Guoqiang Zhou, Yi Fan, Jiachen Shi, Yuyuan Lu, Jun Shen
2022 Applied Sciences  
Generative Adversarial Network (GAN), deemed as a powerful deep-learning-based silver bullet for intelligent data generation, has been widely used in multi-disciplines.  ...  Furthermore, conditional GAN (CGAN) introduces artificial control information on the basis of GAN, which is more practical for many specific fields, though it is mostly used in domain transfer.  ...  Methods Adversarial Loss Content Loss Perception Loss Hidden Variable Loss Category Loss Laplacian Pyramid [28] SRGAN [29] Amortised MAP Inference [30] ESRGAN [14] Pixel-Level Domain Transfer  ... 
doi:10.3390/app12168350 fatcat:4qjfoetexfaevoptzecu5i2jqy

Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models [article]

Sam Bond-Taylor, Adam Leach, Yang Long, Chris G. Willcocks
2021 arXiv   pre-print
Beyond this, generative modelling has numerous direct applications including image synthesis: super-resolution, text-to-image and image-to-image conversion, inpainting, attribute manipulation, pose estimation  ...  By sampling using a finer grid of coordinates, super-resolution beyond resolutions seen during training is possible.  ... 
arXiv:2103.04922v2 fatcat:nivlg3whyjhadhwdl2tsh5yciy

Deep Generative Modelling: A Comparative Review of VAEs, GANs, Normalizing Flows, Energy-Based and Autoregressive Models

Sam Bond-Taylor, Adam Leach, Yang Long, Chris George Willcocks
2021 IEEE Transactions on Pattern Analysis and Machine Intelligence  
Beyond this, generative modelling has numerous direct applications including image synthesis: super-resolution, text-to-image and image-to-image conversion, inpainting, attribute manipulation, pose estimation  ...  By sampling using a finer grid of coordinates, super-resolution beyond resolutions seen during training is possible.  ... 
doi:10.1109/tpami.2021.3116668 pmid:34591756 fatcat:yjpayhmrfnaeziahmrgiyvtxkm

Self-Supervised Radio-Visual Representation Learning for 6G Sensing [article]

Mohammed Alloulah, Akash Deep Singh, Maximilian Arnold
2022 arXiv   pre-print
This indicates that self-supervised learning could be an important enabler for future scalable radio sensing systems.  ...  In future 6G cellular networks, a joint communication and sensing protocol will allow the network to perceive the environment, opening the door for many new applications atop a unified communication-perception  ...  In [35] , the authors propose a super-resolution method called Radar signal Reconstruction using Self-supervision (R2-S2) which improves the angular resolution of a given radar array without increasing  ... 
arXiv:2111.02887v2 fatcat:jr63gi67gvcphaeh7a2ypcvnmi

Top-Down Deep Clustering with Multi-generator GANs [article]

Daniel de Mello, Renato Assunção, Fabricio Murai
2021 arXiv   pre-print
We use this clustered generation to train a classifier for inferring from which generator a given image came from, thus providing a semantically meaningful clustering for the real distribution.  ...  This approach filters out low-level information irrelevant for clustering and has proven remarkably successful for high dimensional data spaces.  ...  Amortised MAP Inference for Image evance for this work are the following: Nvidia RTX 3070 Super-resolution.  ... 
arXiv:2112.03398v2 fatcat:232fy3nryneu3bi4zqaqkoz5zm

The Detailed Science Case for the Maunakea Spectroscopic Explorer: the Composition and Dynamics of the Faint Universe [article]

Alan McConnachie, Carine Babusiaux, Michael Balogh, Simon Driver, Pat Côté, Helene Courtois, Luke Davies, Laura Ferrarese, Sarah Gallagher, Rodrigo Ibata, Nicolas Martin, Aaron Robotham, Kim Venn (+149 others)
2016 arXiv   pre-print
It is an ideal feeder facility for E-ELT, TMT and GMT, and provides the missing link between wide field imaging and small field precision astronomy.  ...  ~ 40000) resolution.  ...  With sufficient SNR, time sampling, velocity resolution, and flux calibration, the velocity-delay map can be inverted to reconstruct the disk image. (Image credit: K.  ... 
arXiv:1606.00043v1 fatcat:2atxpt2u5zabfc5pjhzd5c5a3e

Affective Processes: stochastic modelling of temporal context for emotion and facial expression recognition [article]

Enrique Sanchez and Mani Kumar Tellamekala and Michel Valstar and Georgios Tzimiropoulos
2021 arXiv   pre-print
We validate our approach on four databases, two for Valence and Arousal estimation (SEWA and AffWild2), and two for Action Unit intensity estimation (DISFA and BP4D).  ...  To alleviate these issues, we build upon the framework of Neural Processes to propose a method for apparent emotion recognition with three key novel components: (a) probabilistic contextual representation  ...  Acknowledgements The work by Michel Valstar was part-funded by the National Institute for Health Research.  ... 
arXiv:2103.13372v1 fatcat:auaeihk46vekno2rshypknxqpa
« Previous Showing results 1 — 15 out of 73 results