miguelvr/dropblock
PyTorch library implementing DropBlock, a structured dropout variant for convolutional neural networks that drops contiguous regions of feature maps.
Not currently ranked — collecting fresh signals.
star history
This repository provides a PyTorch implementation of DropBlock, a regularization technique specifically designed for convolutional layers. Unlike standard dropout which is less effective on spatially correlated activation units in conv layers, DropBlock drops contiguous regions of feature maps together, improving regularization. The implementation supports configurable block sizes and includes integration options for skip connections. It targets computer vision models where convolutional networks dominate.
Frequently asked
- What is miguelvr/dropblock?
- PyTorch library implementing DropBlock, a structured dropout variant for convolutional neural networks that drops contiguous regions of feature maps.
- Is dropblock open source?
- Yes — miguelvr/dropblock is open source, released under the MIT license.
- What language is dropblock written in?
- miguelvr/dropblock is primarily written in Python.
- How popular is dropblock?
- miguelvr/dropblock has 594 stars on GitHub.
- Where can I find dropblock?
- miguelvr/dropblock is on GitHub at https://github.com/miguelvr/dropblock.