raghakot/keras-vis

Health v2 · maintenance only

Adoption not part of the score

2991 stars · 635 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-29, confidence not recorded

keras-vis is a high-level Python toolkit for visualizing and debugging trained Keras neural network models. It supports activation maximization, saliency maps, and class activation maps, framed as energy minimization problems with an extensible loss/regularizer interface.

Use cases

• visualize what features a keras CNN layer has learned
• generate saliency maps to explain model predictions
• compute class activation maps for image classifiers
• debug a trained neural network by inspecting activations
• generate images that maximally activate a filter
• understand why a keras image model misclassifies

When to choose

• you maintain a legacy Keras (or Theano/TensorFlow 1.x backend) model and need activation maximization, saliency, or CAM visualizations
• you want a simple, extensible loss/regularizer interface for image backprop experiments

When to avoid

• you use modern TensorFlow 2.x / tf.keras or PyTorch - the project is unmaintained and incompatible with current versions
• you need actively supported interpretability tooling - prefer alternatives like tf-keras-vis, Grad-CAM libraries, or Captum
• you need non-image modality visualizations

Facets

library · maturity abandoned

machine-learning data-visualization deep-learning deep-learning machine-learning data-visualization python keras neural-networks saliency-maps activation-maximization class-activation-maps model-interpretability explainability

Member repositories

For agents

markdown · JSON · MCP: product_card(name="raghakot/keras-vis")