[1805.08974] Do Better ImageNet Models Transfer Better?open searchopen navigation menucontact arXivsubscribe to arXiv mailings

Transfer learning is a cornerstone of computer vision, yet little work has been done to evaluate the relationship between architecture and transfer. An implicit hypothesis in modern computer vision research is that models that perform better on ImageNet necessarily perform better on other vision tasks. However, this hypothesis has never been systematically tested. Here, we compare the performance of 16 classification networks on 12 image classification datasets. We find that, when networks are used as fixed feature extractors or fine-tuned, there is a strong correlation between ImageNet accuracy and transfer accuracy ($r = 0.99$ and $0.96$, respectively). In the former setting, we find that this relationship is very sensitive to the way in which networks are trained on ImageNet; many common forms of regularization slightly improve ImageNet accuracy but yield penultimate layer features that are much worse for transfer learning. Additionally, we find that, on two small fine-grained image

1 mentions: @nzw0301
Keywords: imagenet
Date: 2020/10/16 09:52

Referring Tweets

@nzw0301 @mosko_mule ImageNetに対して汎化してしまえば、他のタスクにもある程度transferableだと思える (ex. t.co/Urvz6jR8ZI )のと、MIのlower bound最大化でほぼ性能を説明するのに失敗してる(t.co/AbIJr1ojlY)ことを考えると悪くないと思うのですが、なにか引っかかる感じでしょうか。

Bookmark Comments

Related Entries

Read more [MIRU2018] Global Average Poolingの特性を用いたAttention Branch Network
2 users, 0 mentions 2018/08/10 09:23
Read more GitHub - fidler-lab/curve-gcn: Official PyTorch code for Curve-GCN (CVPR 2019)
1 users, 0 mentions 2019/07/31 09:47
Read more GitHub - vikasverma1077/manifold_mixup: Code for reproducing Manifold Mixup results (ICML 2019)
1 users, 0 mentions 2019/09/11 03:48
Read more GitHub - justusschock/deep_alignment_network_pytorch: PyTorch Implementation of the Deep Alignment N...
1 users, 0 mentions 2019/10/14 03:48
Read more GitHub - youngwanLEE/CenterMask: CenterMask : Real-Time Anchor-Free Instance Segmentation, in CVPR20...
1 users, 0 mentions 2020/02/25 09:51