huangzehao/resnet.mxnet

resnet.mxnet

This repository contains a MXNet implementation for the ResNet-v2 and ResNeXt algorithms for image classification.

Requirements and Dependencies

• Install a modified MXNet on a machine with CUDA GPU
• Install CUDNN v5 or v6
• Download the ImageNet dataset and creat pass through rec (following tornadomeet's repository but using unchange mode)

Train and Test

• Modify config/cfg.py by your setting
• python train.py
• python test.py

Reference

[1] Kaiming He, et al. "Identity Mappings in Deep Residual Networks" arXiv:1603.05027 (2016).

[2] Xie, Saining, et al. "Aggregated residual transformations for deep neural networks." arXiv:1611.05431 (2016).

[3] Chen, Tianqi, et al. "Mxnet: A flexible and efficient machine learning library for heterogeneous distributed systems." arXiv:1512.01274 (2015).

[4] Torch training code and model provided by facebook, https://github.com/facebookresearch/ResNeXt

[5] MXNet training code provided by tornadomeet, https://github.com/tornadomeet/ResNet