clm_d449eac4Based 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.
arXiv 1905.11946 · cs.LG cs.CV stat.ML
Read on arXiv · arXiv non-exclusive license
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.
clm_d449eac4Based 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.
clm_59e17bd3We demonstrate the effectiveness of this method on scaling up MobileNets and ResNet.
res_c4f881d0In 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.
res_41463a79Our 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.
clm_46eedf21Our EfficientNets significantly outperform other ConvNets.
res_7922b6bfIn 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.
clm_3753e328Based on this observation, we propose a simple yet effective compound scaling method.
res_78f6f2e8For 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)
clm_b58e91d0We demonstrate that our scaling method work well on existing MobileNets (Howard et al., 2017; Sandler et al., 2018) and ResNet (He et al., 2016).
res_40d1173bFigure 1 summarizes the ImageNet performance, where our EfficientNets significantly outperform other ConvNets.
res_9dc24c13In 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.
res_2c8c1ed3Compared 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.
res_2c41d698Besides 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.
res_8fae791bFigure 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)
res_0efbbd1cThe 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)
res_1bcf2631Correspondingly, 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)
res_f740d60fWith deeper (d=2.0) and higher resolution (r=2.0), width scaling achieves much better accuracy under the same FLOPS cost.
res_24aed27dThese 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)
clm_b912d6c1In 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)
res_e35a0041Our scaled EfficientNet models achieve new state-of-theart accuracy for 5 out of 8 datasets, with 9.6x fewer parameters on average.
res_6b93dafeIn 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).
res_25f7f862Figure 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)
res_6ec59a21For example, our EfficientNet-B3 achieves higher accuracy than ResNeXt- 101 (Xie et al., 2017) using 18x fewer FLOPS.
res_94aac8ebTable 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)
res_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)
res_626e7dc6In 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)
res_3822f80cIn 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)
clm_8eb9e72aIn 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)
clm_bbe70226To 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)
clm_12a482bbPowered 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)
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.