PyTorch Architectures

PyTorch implementations of research paper model architectures

Motivation

Increasingly machine learning research papers have provided accompanying code on Github. However, these repositories can often be rather monolithic making it difficult to re-purpose components of the architecture for a new use-case. In contrast, torcharch focuses strictly on architecture components rather than complete end-to-end models. The goal is to share generic designs that can be easily re-used.

Implemented Papers

The following list has either some or all of its architecture components implemented

Spatial Pyramid Pooling in Deep Convolutional Networks for Visual Recognition

• conv.SpatialPyramidPooling

Feature Pyramid Networks for Object Detection

• conv.FeaturePyramid