MearcAll papers

arXiv 1905.11946 · cs.LG cs.CV stat.ML

EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks

Mingxing Tan, Quoc V. Le

Claims were extracted automatically by reducer-1 on 2026-11-12. The paper's license does not allow republishing it, so long claims are shortened to a quotation. Read the full text on arXiv.

No reproduction recordedclm_d449eac4
Based on this observation, we propose a new scaling method that uniformly scales all dimensions of depth/width/resolution using a simple yet highly effective compound coefficient.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_59e17bd3
We demonstrate the effectiveness of this method on scaling up MobileNets and ResNet.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_c4f881d0
In particular, our EfficientNet-B7 achieves state-of-the-art 84.3% top-1 accuracy on ImageNet, while being 8.4x smaller and 6.1x faster on inference than the best existing ConvNet.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_41463a79
Our EfficientNets also transfer well and achieve state-of-the-art accuracy on CIFAR-100 (91.7%), Flowers (98.8%), and 3 other transfer learning datasets, with an order of magnitude fewer parameters.

Abstract · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_46eedf21
Our EfficientNets significantly outperform other ConvNets.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_7922b6bf
In particular, EfficientNet-B7 achieves new state-of-the-art 84.3% top-1 accuracy but being 8.4x smaller and 6.1x faster than GPipe.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_3753e328
Based on this observation, we propose a simple yet effective compound scaling method.

Introduction · page 1 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_78f6f2e8
For example, if we want to use 2N times more computational resources, then we can simply increase the network depth by αN , width byβN , and image size byγN … (excerpt)

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_b58e91d0
We demonstrate that our scaling method work well on existing MobileNets (Howard et al., 2017; Sandler et al., 2018) and ResNet (He et al., 2016).

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_40d1173b
Figure 1 summarizes the ImageNet performance, where our EfficientNets significantly outperform other ConvNets.

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_9dc24c13
In particular, our EfficientNet-B7 surpasses the best existing GPipe accuracy (Huang et al., 2018), but using 8.4x fewer parameters and running 6.1x faster on inference.

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_2c8c1ed3
Compared to the widely used ResNet-50 (He et al., 2016), our EfficientNet-B4 improves the top-1 accuracy from 76.3% to 83.0% (+6.7%) with similar FLOPS.

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_2c41d698
Besides ImageNet, EfficientNets also transfer well and achieve stateof-the-art accuracy on 5 out of 8 widely used datasets, while reducing parameters by up to 21x than existing ConvNets.

Introduction · page 2 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_8fae791b
Figure 3 (right) shows the results of scaling network resolutions, where indeed higher resolutions improve accuracy, but the accuracy gain diminishes for very high resolutions (r = 1.0 denotes resolution … (excerpt)

Scaling Dimensions · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_0efbbd1c
The above analyses lead us to the first observation: Observation 1 – Scaling up any dimension of network width, depth, or resolution improves accuracy, but the accuracy gain diminishes for … (excerpt)

Scaling Dimensions · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_1bcf2631
Correspondingly, we should also increase network width when resolution is higher, in 2In some literature, scaling number of channels is called “depth multiplier”, which means the same as our width … (excerpt)

Compound Scaling · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_f740d60f
With deeper (d=2.0) and higher resolution (r=2.0), width scaling achieves much better accuracy under the same FLOPS cost.

If we only scale network width w without changing · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_24aed27d
These results lead us to the second observation: Observation 2 – In order to pursue better accuracy and efficiency, it is critical to balance all dimensions of network width, depth, … (excerpt)

If we only scale network width w without changing · page 4 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_b912d6c1
In this paper, we propose a newcompound scaling method, which use a compound coefficient φ to uniformly scales network width, depth, and resolution in a principled way: depth:d =αφ width:w … (excerpt)

If we only scale network width w without changing · page 5 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_e35a0041
Our scaled EfficientNet models achieve new state-of-theart accuracy for 5 out of 8 datasets, with 9.6x fewer parameters on average.

Scaling Up MobileNets and ResNets · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_6b93dafe
In particular, our EfficientNet-B7 achieves 84.3% top1 accuracy with 66M parameters and 37B FLOPS, being more accurate but 8.4x smaller than the previous best GPipe (Huang et al., 2018).

Scaling Up MobileNets and ResNets · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_25f7f862
Figure 1 and Figure 5 illustrates the parameters-accuracy and FLOPS-accuracy curve for representative ConvNets, where our scaled EfficientNet models achieve better accuracy with much fewer parameters and FLOPS than other … (excerpt)

Scaling Up MobileNets and ResNets · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_6ec59a21
For example, our EfficientNet-B3 achieves higher accuracy than ResNeXt- 101 (Xie et al., 2017) using 18x fewer FLOPS.

Scaling Up MobileNets and ResNets · page 7 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_94aac8eb
Table 5 shows the transfer learning performance: (1) Compared to public available models, such as NASNet-A (Zoph et al., 2018) and Inception-v4 (Szegedy et al., 2017), our EfficientNet models achieve … (excerpt)

We borrow the same training settings from (Kornblith · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_9d473d80
(2) Compared to stateof-the-art models, including DAT (Ngiam et al., 2018) that dynamically synthesizes training data and GPipe (Huang et al., 2018) that is trained with specialized pipeline parallelism, our … (excerpt)

We borrow the same training settings from (Kornblith · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_626e7dc6
In general, all scaling methods improve accuracy with the cost of more FLOPS, but our compound scaling method can further improve accuracy, by up to 2.5%, than other singledimension scaling … (excerpt)

Discussion · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedres_3822f80c
In order to further understand why our compound scaling method is better than others, Figure 7 compares the class activation map (Zhou et al., 2016) for a few representative models … (excerpt)

Discussion · page 8 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_8eb9e72a
In this paper, we systematically study ConvNet scaling and identify that carefully balancing network width, depth, and resolution is an important but missing piece, preventing us from better accuracy and … (excerpt)

Conclusion · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_bbe70226
To address this issue, we propose a simple and highly effective compound scaling method, which enables us to easily scale up a baseline ConvNet to any target resource constraints in … (excerpt)

Conclusion · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

No reproduction recordedclm_12a482bb
Powered by this compound scaling method, we demonstrate that a mobilesize EfficientNet model can be scaled up very effectively, surpassing state-of-the-art accuracy with an order of magnitude fewer parameters and … (excerpt)

Conclusion · page 9 · Automated extraction, not reviewed · 0 reviewed reproductions

Reproduced one of these claims?

Record your result in the Mearc app on your own machine, then use Publish this result with this paper's id (1905.11946) and the claim id shown next to the claim. Mearc creates a file that holds only what you choose to publish. Add it to the public claim data by pull request; a maintainer other than you reviews it before it counts.