tensorflow/lucid

# Technical Summary Provides feature visualization and attribution techniques via differentiable optimization of image parameterizations, enabling direct visualization of learned neural network representations. Includes a model zoo of 27 pre-imported vision models with a consistent API, plus spatial and channel attribution methods for understanding neuron interactions. Built on TensorFlow 1.x with Jupyter notebooks as primary interface; integrates with pre-trained models from standard computer vision benchmarks.

4,703 stars. No commits in the last 6 months.