ResNet

Deep Residual Learning for Image Recognition

Introduction

Residual Networks, or ResNets, learn residual functions with reference to the layer inputs, instead of learning unreferenced functions. In the mainstream previous works, like VGG, the neural networks are a stack of layers and every layer attempts to fit a desired underlying mapping. In ResNets, a few stacked layers are grouped as a block, and the layers in a block attempts to learn a residual mapping.

Formally, denoting the desired underlying mapping of a block as $\mathcal{H}(x)$, split the underlying mapping into the sum of the identity and the residual mapping as $\mathcal{H}(x) = x + \mathcal{F}(x)$, and let the stacked non-linear layers fit the residual mapping $\mathcal{F}(x)$.

Many works proved this method makes deep neural networks easier to optimize, and can gain accuracy from considerably increased depth. Recently, the residual structure is widely used in various models.

Abstract

Show the paper's abstract

Deeper neural networks are more difficult to train. We present a residual learning framework to ease the training of networks that are substantially deeper than those used previously. We explicitly reformulate the layers as learning residual functions with reference to the layer inputs, instead of learning unreferenced functions. We provide comprehensive empirical evidence showing that these residual networks are easier to optimize, and can gain accuracy from considerably increased depth. On the ImageNet dataset we evaluate residual nets with a depth of up to 152 layers---8x deeper than VGG nets but still having lower complexity. An ensemble of these residual nets achieves 3.57% error on the ImageNet test set. This result won the 1st place on the ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100 and 1000 layers.

The depth of representations is of central importance for many visual recognition tasks. Solely due to our extremely deep representations, we obtain a 28% relative improvement on the COCO object detection dataset. Deep residual nets are foundations of our submissions to ILSVRC & COCO 2015 competitions, where we also won the 1st places on the tasks of ImageNet detection, ImageNet localization, COCO detection, and COCO segmentation.

How to use it?

Predict image

>>> import torch
>>> from mmcls.apis import init_model, inference_model
>>>
>>> model = init_model('configs/resnet/resnet50_8xb32_in1k.py', 'https://download.openmmlab.com/mmclassification/v0/resnet/resnet50_8xb32_in1k_20210831-ea4938fc.pth')
>>> predict = inference_model(model, 'demo/demo.JPEG')
>>> print(predict['pred_class'])
sea snake
>>> print(predict['pred_score'])
0.6649363040924072

Use the model

>>> import torch
>>> from mmcls.apis import init_model
>>>
>>> model = init_model('configs/resnet/resnet50_8xb32_in1k.py', 'https://download.openmmlab.com/mmclassification/v0/resnet/resnet50_8xb32_in1k_20210831-ea4938fc.pth')
>>> inputs = torch.rand(1, 3, 224, 224).to(model.data_preprocessor.device)
>>> # To get classification scores.
>>> out = model(inputs)
>>> print(out.shape)
torch.Size([1, 1000])
>>> # To extract features.
>>> outs = model.extract_feat(inputs)
>>> print(outs[0].shape)
torch.Size([1, 2048])

Train/Test Command

Place the ImageNet dataset to the data/imagenet/ directory, or prepare datasets according to the docs.

Train:

python tools/train.py configs/resnet/resnet50_8xb32_in1k.py

Test:

python tools/test.py configs/resnet/resnet50_8xb32_in1k.py https://download.openmmlab.com/mmclassification/v0/resnet/resnet50_b16x8_cifar10_20210528-f54bfad9.pth

For more configurable parameters, please refer to the API.

Results and models

The pre-trained models on ImageNet-21k are used to fine-tune, and therefore don't have evaluation results.

Model	resolution	Params(M)	Flops(G)	Download
ResNet-50-mill	224x224	86.74	15.14	model

The "mill" means using the mutil-label pretrain weight from ImageNet-21K Pretraining for the Masses.

Cifar10

Model	Params(M)	Flops(G)	Top-1 (%)	Top-5 (%)	Config	Download
ResNet-18	11.17	0.56	94.82	99.87	config	model \| log
ResNet-34	21.28	1.16	95.34	99.87	config	model \| log
ResNet-50	23.52	1.31	95.55	99.91	config	model \| log
ResNet-101	42.51	2.52	95.58	99.87	config	model \| log
ResNet-152	58.16	3.74	95.76	99.89	config	model \| log

Cifar100

Model	Params(M)	Flops(G)	Top-1 (%)	Top-5 (%)	Config	Download
ResNet-50	23.71	1.31	79.90	95.19	config	model \| log

ImageNet-1k

Model	Params(M)	Flops(G)	Top-1 (%)	Top-5 (%)	Config	Download
ResNet-18	11.69	1.82	69.90	89.43	config	model \| log
ResNet-34	21.8	3.68	73.62	91.59	config	model \| log
ResNet-50	25.56	4.12	76.55	93.06	config	model \| log
ResNet-101	44.55	7.85	77.97	94.06	config	model \| log
ResNet-152	60.19	11.58	78.48	94.13	config	model \| log
ResNetV1C-50	25.58	4.36	77.01	93.58	config	model \| log
ResNetV1C-101	44.57	8.09	78.30	94.27	config	model \| log
ResNetV1C-152	60.21	11.82	78.76	94.41	config	model \| log
ResNetV1D-50	25.58	4.36	77.54	93.57	config	model \| log
ResNetV1D-101	44.57	8.09	78.93	94.48	config	model \| log
ResNetV1D-152	60.21	11.82	79.41	94.70	config	model \| log
ResNet-50 (fp16)	25.56	4.12	76.30	93.07	config	model \| log
Wide-ResNet-50*	68.88	11.44	78.48	94.08	config	model
Wide-ResNet-101*	126.89	22.81	78.84	94.28	config	model
ResNet-50 (rsb-a1)	25.56	4.12	80.12	94.78	config	model \| log
ResNet-50 (rsb-a2)	25.56	4.12	79.55	94.37	config	model \| log
ResNet-50 (rsb-a3)	25.56	4.12	78.30	93.80	config	model \| log

The "rsb" means using the training settings from ResNet strikes back: An improved training procedure in timm.

Models with * are converted from the official repo. The config files of these models are only for validation. We don't ensure these config files' training accuracy and welcome you to contribute your reproduction results.

CUB-200-2011

Model	Pretrain	resolution	Params(M)	Flops(G)	Top-1 (%)	Config	Download
ResNet-50	ImageNet-21k-mill	448x448	23.92	16.48	88.45	config	model \| log

Citation

@inproceedings{he2016deep,
  title={Deep residual learning for image recognition},
  author={He, Kaiming and Zhang, Xiangyu and Ren, Shaoqing and Sun, Jian},
  booktitle={Proceedings of the IEEE conference on computer vision and pattern recognition},
  pages={770--778},
  year={2016}
}