By Kaiming He, Xiangyu Zhang, Shaoqing Ren, Jian Sun.

Microsoft Research Asia (MSRA).

0. Introduction
1. Disclaimer and known issues
2. Models
3. Results
4. Third-party re-implementations

This repository contains the original models (ResNet-50, ResNet-101, and ResNet-152) described in the paper "Deep Residual Learning for Image Recognition" (http://arxiv.org/abs/1512.03385). These models are those used in [ILSVRC] (http://image-net.org/challenges/LSVRC/2015/) and COCO 2015 competitions, which won the 1st places in: ImageNet classification, ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

Note - Check re-implementations with training code and models from Facebook AI Research (FAIR)! -- blog, code

If you use these models in your research, please cite:

@article{He2015,
	author = {Kaiming He and Xiangyu Zhang and Shaoqing Ren and Jian Sun},
	title = {Deep Residual Learning for Image Recognition},
	journal = {arXiv preprint arXiv:1512.03385},
	year = {2015}
}

0. These models are converted from our own implementation to a recent version of Caffe (2016/2/3, b590f1d). The numerical results using this code are as in the tables below.
1. These models are for the usage of testing or fine-tuning.
2. These models were not trained using this version of Caffe.
3. If you want to train these models using this version of Caffe without modifications, please notice that:
   • GPU memory might be insufficient for extremely deep models.
   • Changes of mini-batch size should impact accuracy (we use a mini-batch of 256 images on 8 GPUs, that is, 32 images per GPU).
   • Implementation of data augmentation might be different (see our paper about the data augmentation we used).
   • We randomly shuffle data at the beginning of every epoch.
   • There might be some other untested issues.
4. In our BN layers, the provided mean and variance are strictly computed using average (not moving average) on a sufficiently large training batch after the training procedure. The numerical results are very stable (variation of val error < 0.1%). Using moving average might lead to different results.
5. In the BN paper, the BN layer learns gamma/beta. To implement BN in this version of Caffe, we use its provided "batch_norm_layer" (which has no gamma/beta learned) followed by "scale_layer" (which learns gamma/beta).
6. We use Caffe's implementation of SGD with momentum: v := momentum*v + lr*g. If you want to port these models to other libraries (e.g., Torch, CNTK), please pay careful attention to the possibly different implementation of SGD with momentum: v := momentum*v + (1-momentum)*lr*g, which changes the effective learning rates.

0. Visualizations of network structures (tools from ethereon):

   • [ResNet-50] (http://ethereon.github.io/netscope/#/gist/db945b393d40bfa26006)
   • [ResNet-101] (http://ethereon.github.io/netscope/#/gist/b21e2aae116dc1ac7b50)
   • [ResNet-152] (http://ethereon.github.io/netscope/#/gist/d38f3e6091952b45198b)

1. Model files:

   • MSR download: [link] (http://research.microsoft.com/en-us/um/people/kahe/resnet/models.zip)
   • OneDrive download: link

0. Curves on ImageNet (solid lines: 1-crop val error; dashed lines: training error):

1. 1-crop validation error on ImageNet (center 224x224 crop from resized image with shorter side=256):

   model	top-1	top-5
   VGG-16	28.5%	9.9%
   ResNet-50	24.7%	7.8%
   ResNet-101	23.6%	7.1%
   ResNet-152	23.0%	6.7%

2. 10-crop validation error on ImageNet (averaging softmax scores of 10 224x224 crops from resized image with shorter side=256), the same as those in the paper:

   model	top-1	top-5
   ResNet-50	22.9%	6.7%
   ResNet-101	21.8%	6.1%
   ResNet-152	21.4%	5.7%

Deep residual networks are very easy to implement and train. We recommend to see also the following third-party re-implementations and extensions:

0. By Facebook AI Research (FAIR), with training code in Torch and pre-trained ResNet-18/34/50/101 models for ImageNet: blog, code
1. Torch, CIFAR-10, with ResNet-20 to ResNet-110, training code, and curves: code
2. Lasagne, CIFAR-10, with ResNet-32 and ResNet-56 and training code: code
3. Neon, CIFAR-10, with pre-trained ResNet-32 to ResNet-110 models, training code, and curves: code
4. Torch, MNIST, 100 layers: blog, code
5. A winning entry in Kaggle's right whale recognition challenge: blog, code
6. Neon, Place2 (mini), 40 layers: blog, code

In addition, this [code] (https://github.com/ry/tensorflow-resnet) by Ryan Dahl helps to convert the pre-trained models to TensorFlow.