4 repositorios
Layers that inject learnable spatial information into image patch sequences.
Distinct from Positional Encoding Techniques: Focuses on the implementation of positional embeddings for image patches, whereas the parent covers general positional encoding techniques.
Explore 4 awesome GitHub repositories matching artificial intelligence & ml · Positional Embedding Layers. Refine with filters or upvote what's useful.
This project is an educational platform and research toolkit designed to teach deep learning through a combination of mathematical theory, visual diagrams, and executable code. It provides a comprehensive environment for building, training, and evaluating neural networks, grounding complex concepts in interactive computational notebooks that allow for hands-on experimentation. The framework distinguishes itself by interleaving theoretical foundations—including linear algebra, calculus, and probability—with practical implementations across multiple industry-standard libraries. It supports flex
Adds learnable vectors to image patch representations to preserve spatial information in transformer encoders.
PlugNPlay-Modules is a collection of reusable PyTorch computer vision modules and deep learning architectural components. It provides a library of standardized building blocks for constructing neural networks, focusing on attention mechanisms, signal processing layers, and feature fusion modules. The project is distinguished by its extensive variety of attention primitives, covering spatial, channel, and temporal weighting, as well as specialized variants like deformable, frequency-enhanced, and linear-complexity attention. It also implements advanced signal processing tools within the neural
Creates sine and cosine positional encodings to provide spatial or sequential awareness to network layers.
DeiT es un framework de vision transformer para PyTorch diseñado para la clasificación de imágenes. Implementa una arquitectura basada en transformers que procesa imágenes como secuencias de parches aplanados utilizando capas de auto-atención (self-attention) y modelado de secuencias consciente de la posición en lugar de filtros convolucionales. El proyecto se centra en el entrenamiento eficiente en datos a través de un framework de destilación de conocimiento. Este sistema permite que un modelo estudiante imite las etiquetas blandas (soft labels) de un modelo profesor de alto rendimiento para mejorar la precisión y la generalización, particularmente cuando se entrena con conjuntos de datos más pequeños. La librería cubre el ciclo de vida completo de desarrollo, incluyendo el entrenamiento de clasificación de imágenes, la optimización de la pérdida de entropía cruzada y el despliegue de pesos preentrenados para inferencia. También incluye una herramienta de benchmarking para evaluar el rendimiento y la precisión del modelo frente a conjuntos de datos estándar.
Injects learnable positional embeddings into image patch sequences to preserve spatial arrangement.
Neuraloperator is a library for learning mappings between infinite-dimensional function spaces, serving as a tool to accelerate physics simulations and partial differential equation solving. It implements resolution-invariant models and spectral neural networks that can produce consistent predictions regardless of the input grid resolution or spatial discretization. The framework incorporates physics-informed neural networks that enforce physical constraints and differential equations through specialized loss functions. It utilizes Fourier transforms and spectral projections to process multid
Adds spectral or grid-based positional information to coordinate inputs as additional data channels.