kevinzakka/spatial-transformer-network

Health v2 · maintenance only

Flags: no_releases

Adoption not part of the score

1005 stars · 266 forks observed · 2026-09-03

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

A TensorFlow implementation of Spatial Transformer Networks, a differentiable module that can be inserted into ConvNet architectures to add geometric invariance via learned affine transformations. It is packaged as a pip-installable `stn` library with a blog tutorial series explaining the underlying concepts.

Use cases

• add a spatial transformer layer to a tensorflow convnet
• make a cnn invariant to rotation and translation
• learn affine transformations of feature maps without extra supervision
• crop and scale image regions with a differentiable attention module
• implement spatial transformer networks from the jaderberg paper
• upsample or downsample feature maps inside a neural network

When to choose

• you are building CNNs in TensorFlow 1.x and need a differentiable spatial transformer layer
• you want a minimal, well-documented STN implementation to learn from or extend
• you need learned cropping, translation, or scaling as part of a model

When to avoid

• you use PyTorch or another framework instead of TensorFlow
• you need actively maintained code compatible with modern TensorFlow 2.x
• you need production-grade, heavily tested computer vision tooling

Facets

library · maturity maintenance

machine-learning deep-learning image-processing deep-learning computer-vision machine-learning python spatial-transformer-network tensorflow attention affine-transformation convnet neural-network-layers

Member repositories

For agents

markdown · JSON · MCP: product_card(name="kevinzakka/spatial-transformer-network")