Vision Research Toolkit by PyTorch

Why should we build a new framework every time starting a new research?

Welcome to VRT by PyTorch!

• VRT is a high robust PyTorch-based framework for deep learning in computer vision, which aimed at easy-to-use, high efficiency, clean code and research.
• The dataset in VRT is efficient enough to support all kind of task (seg, det, class, etc..)
• All the parameters in VRT can be controlled in one config file.
• VRT support single GPU and multi GPU, but do not support CPU.

Acknowledgement

• The idea was inspired by the project torchcv.
• The base framework is heavily influenced by maskrcnn-benchmark, which is a really cool implementation of PyTorch.