Skip to content

Latest commit

 

History

History
14 lines (11 loc) · 508 Bytes

README.md

File metadata and controls

14 lines (11 loc) · 508 Bytes

A Faster-RCNN pytorch implementation

The goal was to implement a Faster RCNN model from https://arxiv.org/abs/1506.01497 . In a simple, reader friendly way. The model is not optimized.

Supports Resnet50 and Vgg16.

TODO:

  • Generic interface for dataset (right now it's limited to one of my non-public datasets, under VOC format)
  • Remove anchors out of bonds
  • Improve negative proposals sampling for fastrcnn loss
  • Optimize anchors and ROI operations
  • Switch to a more pytorch syntax, with less numpy