9.2 KiB

Raw Blame History

SimSiam

Exploring Simple Siamese Representation Learning

Siamese networks have become a common structure in various recent models for unsupervised visual representation learning. These models maximize the similarity between two augmentations of one image, subject to certain conditions for avoiding collapsing solutions. In this paper, we report surprising empirical results that simple Siamese networks can learn meaningful representations even using none of the following: (i) negative sample pairs, (ii) large batches, (iii) momentum encoders. Our experiments show that collapsing solutions do exist for the loss and structure, but a stop-gradient operation plays an essential role in preventing collapsing. We provide a hypothesis on the implication of stop-gradient, and further show proof-of-concept experiments verifying it. Our “SimSiam” method achieves competitive results on ImageNet and downstream tasks. We hope this simple baseline will motivate people to rethink the roles of Siamese architectures for unsupervised representation learning.

Citation

@inproceedings{chen2021exploring,
  title={Exploring simple siamese representation learning},
  author={Chen, Xinlei and He, Kaiming},
  booktitle={CVPR},
  year={2021}
}

Models and Benchmarks

Back to model_zoo.md

In this page, we provide benchmarks as much as possible to evaluate our pre-trained models. If not mentioned, all models were trained on ImageNet1k dataset.

VOC SVM / Low-shot SVM

The Best Layer indicates that the best results are obtained from which layers feature map. For example, if the Best Layer is feature3, its best result is obtained from the second stage of ResNet (1 for stem layer, 2-5 for 4 stage layers).

Besides, k=1 to 96 indicates the hyper-parameter of Low-shot SVM.

Model	Config	Best Layer	SVM	k=1	k=2	k=4	k=8	k=16	k=32	k=64	k=96
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

ImageNet Linear Evaluation

The Feature1 - Feature5 don't have the GlobalAveragePooling, the feature map is pooled to the specific dimensions and then follows a Linear layer to do the classification. Please refer to resnet50_mhead_8xb32-steplr-90e.py for details of config.

The AvgPool result is obtained from Linear Evaluation with GlobalAveragePooling. Please refer to file name for details of config.

Model	Config	Feature1	Feature2	Feature3	Feature4	Feature5	AvgPool
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

iNaturalist2018 Linear Evaluation

Please refer to resnet50_mhead_8xb32-steplr-84e_inat18.py and file name for details of config.

Model	Config	Feature1	Feature2	Feature3	Feature4	Feature5	AvgPool
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

Places205 Linear Evaluation

Please refer to resnet50_mhead_8xb32-steplr-28e_places205.py and file name for details of config.

Model	Config	Feature1	Feature2	Feature3	Feature4	Feature5	AvgPool
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

Semi-Supervised Classification

In this benchmark, the necks or heads are removed and only the backbone CNN is evaluated by appending a linear classification head. All parameters are fine-tuned.
When training with 1% ImageNet, we find hyper-parameters especially the learning rate greatly influence the performance. Hence, we prepare a list of settings with the base learning rate from {0.001, 0.01, 0.1} and the learning rate multiplier for the head from {1, 10, 100}. We choose the best performing setting for each method. The setting of parameters are indicated in the file name. The learning rate is indicated like 1e-1, 1e-2, 1e-3 and the learning rate multiplier is indicated like head1, head10, head100.
Please use --deterministic in this benchmark.

Please refer to the directories configs/benchmarks/classification/imagenet/imagenet_1percent/ of 1% data and configs/benchmarks/classification/imagenet/imagenet_10percent/ 10% data for details.

Model	Pretrain Config	Fine-tuned Config	Top-1 (%)	Top-5 (%)
model	resnet50_8xb32-accum16-coslr-200e

Detection

The detection benchmarks includes 2 downstream task datasets, Pascal VOC 2007 + 2012 and COCO2017. This benchmark follows the evluation protocols set up by MoCo.

Pascal VOC 2007 + 2012

Please refer to faster_rcnn_r50_c4_mstrain_24k.py for details of config.

Model	Config	mAP	AP50
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

COCO2017

Please refer to mask_rcnn_r50_fpn_mstrain_1x.py for details of config.

Model	Config	mAP(Box)	AP50(Box)	AP75(Box)	mAP(Mask)	AP50(Mask)	AP75(Mask)
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

Segmentation

The segmentation benchmarks includes 2 downstream task datasets, Cityscapes and Pascal VOC 2012 + Aug. It follows the evluation protocols set up by MMSegmentation.

Pascal VOC 2012 + Aug

Please refer to file for details of config.

Model	Config	mIOU
model	resnet50_8xb32-coslr-100e
model	resnet50_8xb32-coslr-200e

Cityscapes