Search Results for "action"

Nano-Neuron is a didactic project that reduces the idea of a neuron to a handful of tiny JavaScript functions so learners can see “learning” in action without heavy frameworks. It demonstrates how a scalar input can be linearly transformed with a weight and bias, then adjusted via gradient updates to fit a simple mapping such as Celsius-to-Fahrenheit conversion. The code emphasizes readability over performance, inviting you to step through calculations and watch parameters converge. ...

Kinetics-I3D, developed by Google DeepMind, provides trained models and implementation code for the Inflated 3D ConvNet (I3D) architecture introduced in the paper “Quo Vadis, Action Recognition? A New Model and the Kinetics Dataset” (CVPR 2017). The I3D model extends the 2D convolutional structure of Inception-v1 into 3D, allowing it to capture spatial and temporal information from videos for action recognition. This repository includes pretrained I3D models on the Kinetics dataset, with both RGB and optical flow input streams. ...

...Non-local blocks compute attention-like responses across all positions in space-time, allowing a feature at one frame and location to aggregate information from distant frames and regions. This formulation improves action recognition and spatiotemporal reasoning, especially for classes requiring context beyond short temporal windows. The repo provides training recipes and models for standard datasets, as well as ablations that show how many non-local blocks to insert and at which stages. Efficient implementations keep memory and compute manageable so the blocks can be added without rewriting the entire backbone. ...