Pixel discretization fails against advanced adversarial attacks.
problem Understanding limitations of pixel discretization against adversarial attacks.
method Study of pixel discretization defense methods, including variants.
result Pixel discretization is ineffective against strong adversarial attacks.
Improved PixelCNN model with faster training and better performance.
problem Generating high-quality images with tractable likelihood.
method Discretized logistic mixture likelihood, pixel conditioning, downsampling, short-cut connections, dropout regularization.
result State-of-the-art log likelihood results on CIFAR-10.
Random Gaussian noise and pixel discretization improve image classifier robustness.
problem Whitebox adversarial attacks decrease classifier accuracy.
method Inject random Gaussian noise, discretize pixels, and use any classifier.
result Reduces KL divergence and lower bound on classifier accuracy.
We describe an algorithm that associates to each positive real number r and each finite collection Cr of planar pixels of size r a planar piecewise linear set Sr with the following additional property: if Cr is the collection of pixels of size r that touch a given compact semialgebraic set S, then the …
A method trains discrete autoencoders efficiently.
problem Training models with discrete latent variables is difficult.
method Variational autoencoder framework with backpropagation through discrete variables.
result Efficiently learns class and realization of objects from unsupervised data.
Proposes a new model for unsupervised clustering with latent variables.
problem The challenge of unsupervised clustering in machine learning.
method Clustered Generator Model with continuous and discrete latent variables.
result Achieves competitive unsupervised clustering accuracy and disentangled latent representations.
VADD enhances discrete diffusion models by capturing inter-dimensional correlations, improving sample quality.
problem Limited modeling of inter-dimensional dependencies in MDMs degrades performance with few denoising steps.
method Introduces an auxiliary recognition model for latent variable modeling, enabling stable training via variational lower bounds maximization and amortized inference.
result VADD consistently outperforms MDM baselines in sample quality with few denoising steps.
Discretizing input space improves DLN robustness against adversarial attacks.
problem Improving machine learning models' resistance to adversarial attacks.
method Input discretization and Binary Neural Networks (BNNs).
result 2-bit input discretization significantly enhances adversarial robustness with minimal accuracy loss.
Bayesian method improves segmentation accuracy with noisy labels.
problem Annotation errors in semantic segmentation due to mislabeling and spatial correlations.
method Approximate Bayesian estimation with spatially correlated discrete distributions and variational inference.
result The method achieves performance comparable to clean labels under moderate noise levels.
We improve stochastic gradient estimators for large categories using Rao-Blackwellization.
problem Computing gradients over large or infinite categories.
method Rao-Blackwellization to reduce variance of stochastic gradient estimators.
result Improves performance on semi-supervised classification and pixel attention tasks.
In blind hyperspectral unmixing (HU), the pure-pixel assumption is well-known to be powerful in enabling simple and effective blind HU solutions. However, the pure-pixel assumption is not always satisfied in an exact sense, especially for scenarios where pixels are heavily mixed. In the no pure-pixel case, a good blind…
Paper proposes a method to speed up discrete diffusion models by distilling many steps into few.
problem Challenges in capturing dependencies between elements in discrete diffusion models.
method Proposes 'mixture' models and loss functions to distill many sampling steps into few.
result Effective in distilling pretrained discrete diffusion models across image and language domains.
A deep learning system classifies hyperspectral images using denoising autoencoders and pixel mixtures.
problem Hyperspectral image segmentation and classification challenges.
method Multiple class-based denoising autoencoders, mixed pixel training augmentation, and morphological operations.
result High performance on the Salinas dataset, verified by deep neural network and morphological hole-filling.
Any subset of the plane can be approximated by a set of square pixels. This transition from a shape to its pixelation is rather brutal since it destroys geometric and topological information about the shape. Using a technique inspired by Morse Theory, we algorithmically produce a PL approximation of the original shape …
Simplified masked diffusion models improve discrete data generation.
problem Complex model formulations and unclear relationships hinder discrete data generative modeling.
method Developed a simple and general framework for masked diffusion models.
result Models trained on OpenWebText surpass prior diffusion language models and outperform autoregressive models.
SemGANs generate pixel-level accurate semantic images.
problem Generating semantic images with pixel-level accuracy.
method Semantic Generative Adversarial Networks (SemGANs).
result SemGANs outperform standard GANs in semantic image generation tasks.
Deep learning improves PS pixel selection in SAR interferometry.
problem Selecting persistent scatterer pixels for geophysical parameter estimation in multi-temporal SAR interferometry.
method Proposed two deep learning architectures: CNN-ISS and CLSTM-ISS trained on phase history to classify PS and non-PS pixels.
result CLSTM-ISS outperforms conventional methods in PS pixel selection and classification accuracy.
New method uses image-level and pixel-level annotations for brain tumor segmentation.
problem Challenges in obtaining pixel-level annotations for brain tumor segmentation.
method Proposes a learning-based framework that combines both pixel- and image-level annotations.
result Method's performance in segmentation quality is competitive with traditional fully-supervised approach.
New RL method learns control policies from pixel data efficiently.
problem Data-efficient reinforcement learning in high-dimensional image spaces.
method Deep dynamical model for image feature learning and predictive control.
result Data-efficient learning of closed-loop control policies from pixel information.
Pixel-space diffusion models outperform latent models on high-resolution image synthesis.
problem Efficiency and quality trade-off in high-resolution image synthesis.
method Sigmoid loss-weighting, simplified architecture, and resolution scaling.
result Achieved 1.5 FID on ImageNet512, new SOTA results on other datasets.
A method for user-controlled semantic image filling.
problem Generating coherent images with user-specified semantics.
method Deep generative model combining encoder, latent variables, and PixelCNN.
result User can control the inpainting of unobserved pixels while maintaining semantic coherence.
One pixel can significantly alter deep neural network outputs, revealing propagation patterns and vulnerability hotspots.
problem Understanding how a single pixel modification affects deep neural networks.
method Propagation Maps and locality analysis to visualize and understand the impact of pixel modifications.
result One pixel modifications can propagate through deep networks, affecting the final output and revealing vulnerability patterns.
Improves spline quality and accuracy in computational microscopy.
problem Detecting slender, overlapping structures in microscopy images.
method Differentiable rendering approach for spline refinement.
result Achieves high reliability and sub-pixel accuracy.
A new method for adversarial attacks using physical parameters like lighting and geometry.
problem Vulnerability of machine learning image classifiers to adversarial attacks.
method Directly perturbs physical parameters (lighting and geometry) instead of pixel colors, using a differentiable renderer.
result Proposes parametric norm-balls for evaluating adversarial attacks, enabling physically-based attacks.
A new SVM method classifies hyperspectral images using pixel distributions and random Fourier features.
problem Pixel classification in hyperspectral images.
method Representing pixel distributions in RKHS, combining with spectral information, and using random Fourier features to reduce complexity.
result Strong performance on hyperspectral data, outperforming state-of-the-art methods.
Hierarchical randomized smoothing improves model robustness for complex data.
problem Certifying robustness on complex data (e.g. images, graphs) is challenging.
method Add random noise to a randomly selected subset of entities in a hierarchical manner.
result Hierarchical randomized smoothing yields stronger robustness guarantees with high accuracy.
New methods for hyperspectral unmixing handle intra-class variability.
problem Intra-class variability in hyperspectral images.
method Inertia-constrained Pixel-by-pixel NMF (IP-NMF) for handling variability.
result IP-NMF outperforms state-of-the-art methods in real data.
Evolutionary algorithm finds optimal pixel perturbations to improve neural network generalization.
problem Minimal data corruption by pixel modifications causes overfitting in neural networks.
method Evolutionary algorithm with a novel cost function to maximize generalization gap and domain divergence.
result Method outperforms previous pixel-based data distribution shift methods on CNNs.
ALIEN detects multiple small objects and estimates their pixel locations and features.
problem Detecting and localizing multiple small objects in a scene.
method Deep-learning network (ALIEN) that performs simultaneous x,y pixel estimation and feature extraction in a single forward pass.
result Efficient detection and localization of hundreds to thousands of small objects.
Topology-GS improves 3D GS for better structural and feature integrity.
problem Compromised pixel-level and feature-level integrity in 3D GS.
method Incorporates Local Persistent Voronoi Interpolation (LPVI) and PersLoss based on persistent homology.
result Topology-GS outperforms existing methods in PSNR, SSIM, and LPIPS metrics.
Probabilistic inpainting learns multiple plausible images from missing data.
problem Generating multiple plausible images from missing data in images.
method Building a PixelCNN model that learns a distribution of images conditioned on visible pixels.
result The method produces diverse and realistic inpaintings.
Study shows various pixel p-norm measures do not match human perception of adversarial attacks.
problem Understanding human perception of adversarial attacks on image classification systems.
method Performed a behavioral study comparing different p-norm measures and alternative metrics.
result Human perception of adversarial attacks does not align with pixel p-norm measures and other metrics.
DEceit constructs effective universal pixel-restricted perturbations for deep image classifiers.
problem Creating effective universal pixel-restricted perturbations for deep neural networks.
method DEceit algorithm for black-box feedback, targeting 10% of pixels in images.
result Perturbing only 10% of pixels achieves high Fooling Rate and visual similarity.
Simple image augmentation boosts deep RL from pixels.
problem Training deep RL models from raw pixel inputs is challenging.
method Applying standard computer vision data augmentation to reinforcement learning.
result Our method significantly improves SAC's performance on DeepMind control suite.
Pixel-wise classification, where each pixel is assigned to a predefined class, is one of the most important procedures in hyperspectral image (HSI) analysis. By representing a test pixel as a linear combination of a small subset of labeled pixels, a sparse representation classifier (SRC) gives rather plausible results …
SRTC model for background/foreground separation with missing pixels.
problem Background/foreground separation with missing pixels in videos.
method Smooth robust tensor completion (SRTC) model with tensor proximal alternating minimization (tenPAM).
result Global convergence guarantee for the proposed algorithm.
DreamerV2 learns Atari game behaviors from a world model, achieving human-level performance.
problem Learning complex behaviors in Atari games from limited data.
method DreamerV2 uses a world model with discrete representations to predict behaviors in a compact latent space.
result Achieves human-level performance on 55 Atari tasks.
New NAM model improves image processing speed and accuracy.
problem Time-consuming pixel-by-pixel analysis of large images.
method Integrates Neuronal Attention model (NAM) to focus on small regions.
result Efficiently processes large images with improved speed and accuracy.
New method certifies robustness to sparse adversarial attacks.
problem Ensuring robustness of classifiers to sparse adversarial attacks.
method Randomized ablation of input features to provide robustness guarantees.
result Certifies robustness to sparse adversarial attacks with up to 31 pixels of distortion.
Hierarchical autoregressive models improve image quality by learning abstract representations.
problem Local structure bias in autoregressive models leads to lack of large-scale coherence in generated images.
method Propose two methods to learn discrete representations of images that abstract away local detail, and train autoregressive priors on these representations.
result Hierarchical autoregressive models produce high-fidelity reconstructions and realistic images with large-scale coherence.
One pixel attack can fool deep neural networks, showing their vulnerability.
problem Vulnerability of deep neural networks to small perturbations in input.
method Proposes a novel method using differential evolution (DE) for generating one-pixel adversarial perturbations.
result 67.97% of natural images and 16.04% of ImageNet images can be fooled by modifying just one pixel.
Data-efficient learning in continuous state-action spaces using very high-dimensional observations remains a key challenge in developing fully autonomous systems. In this paper, we consider one instance of this challenge, the pixels to torques problem, where an agent must learn a closed-loop control policy from pixel i…
Detects object edges and assigns class labels without pixel-level annotations.
problem Semantic boundary and edge detection with image-level labels.
method Proposes a novel strategy to perform edge detection and class assignment using whole image neural nets and backpropagation.
result High pixel-wise scores indicate semantic boundary locations, suggesting edge labels are not needed during training.
This paper considers a recently emerged hyperspectral unmixing formulation based on sparse regression of a self-dictionary multiple measurement vector (SD-MMV) model, wherein the measured hyperspectral pixels are used as the dictionary. Operating under the pure pixel assumption, this SD-MMV formalism is special in that…
Simple CNN achieves page segmentation for historical documents.
problem Page segmentation of handwritten historical document images.
method Proposes a simple CNN architecture trained on raw image pixels for pixel labeling.
result Simple CNN achieves competitive results compared to other deep architectures.
BriarPatches obscure sensitive attributes to achieve demographic parity.
problem Achieving demographic parity in model predictions.
method Pixel-space interventions that obscure sensitive attributes from classifier representations.
result BriarPatches push downstream predictors towards demographic parity.
Weak supervision enables learning causal representations from unstructured data.
problem Learning high-level causal representations from unstructured data like images.
method Weakly supervised setting with paired samples before and after interventions. Implicit latent causal models using variational autoencoders.
result Models can reliably identify causal structure and disentangle causal variables.
Scene understanding remains a significant challenge in the computer vision community. The visual psychophysics literature has demonstrated the importance of interdependence among parts of the scene. Yet, the majority of methods in computer vision remain local. Pictorial structures have arisen as a fundamental parts-bas…