Search Results for "core"

This repository provides a from-scratch, minimalist implementation of the Vision Transformer (ViT) in PyTorch, focusing on the core architectural pieces needed for image classification. It breaks down the model into patch embedding, positional encoding, multi-head self-attention, feed-forward blocks, and a classification head so you can understand each component in isolation. The code is intentionally compact and modular, which makes it easy to tinker with hyperparameters, depth, width, and attention dimensions. ...

SAM2 is a next-generation version of the Segment Anything Model (SAM), designed to improve performance, generalization, and efficiency in promptable image segmentation tasks. It retains the core promptable interface—accepting points, boxes, or masks—but incorporates architectural and training enhancements to produce higher-fidelity masks, better boundary adherence, and robustness to complex scenes. The updated model is optimized for faster inference and lower memory use, enabling real-time interactivity even on larger images or constrained hardware. ...

Hiera is a hierarchical vision transformer designed to be fast, simple, and strong across image and video recognition tasks. The core idea is to use straightforward hierarchical attention with a minimal set of architectural “bells and whistles,” achieving competitive or superior accuracy while being markedly faster at inference and often faster to train. The repository provides installation options (from source or Torch Hub), a model zoo with pre-trained checkpoints, and code for evaluation and fine-tuning on standard benchmarks. ...