{ "id": "epsje5erax", "slug": "efficientnet", "title": "EfficientNet", "updated": "2026-09-30", "topic_path": [ { "id": "technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/topics/technology" }, { "id": "technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing" }, { "id": "technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai" }, { "id": "technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.ml_computation" }, { "id": "technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep" }, { "id": "technology.computing.ai.neural_deep.alpha-a-to-h", "label": "Titles A to H", "api_url": "https://www.edgepedia.org/api/v1/topics/technology.computing.ai.neural_deep.alpha-a-to-h" } ], "geo": [ { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Non-geographic · 2001 to 2020: Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep", "path": [ { "id": "geo.nongeo", "label": "Non-geographic", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo" }, { "id": "geo.nongeo.t2001", "label": "Non-geographic · 2001 to 2020", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001" }, { "id": "geo.nongeo.t2001.technology", "label": "Technology and the built world", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology" }, { "id": "geo.nongeo.t2001.technology.computing", "label": "Computing and digital systems", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing" }, { "id": "geo.nongeo.t2001.technology.computing.ai", "label": "Artificial intelligence and data", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai" }, { "id": "geo.nongeo.t2001.technology.computing.ai.ml_computation", "label": "Machine learning and neural computation", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.ml_computation" }, { "id": "geo.nongeo.t2001.technology.computing.ai.neural_deep", "label": "Neural networks and deep learning", "api_url": "https://www.edgepedia.org/api/v1/geo/geo.nongeo.t2001.technology.computing.ai.neural_deep" } ] } ], "excerpt": "EfficientNet is a family of convolutional neural networks for image classification that scales depth, width, and resolution together, reported by Mingxing Tan and Quoc V. Le in 2019.", "snippet": "EfficientNet is a family of convolutional neural networks for image classification that scales depth, width, and resolution together, reported by Mingxing Tan and Quoc V. Le in 2019.", "node": "technology.computing.ai.neural_deep.alpha-a-to-h", "markdown": "# EfficientNet\n\nEfficientNet is a family of convolutional neural network architectures for image classification that scales network depth, width, and input resolution together under a single compound coefficient, improving accuracy per unit of computation where earlier networks scaled one dimension at a time. It was reported by Mingxing Tan and Quoc V. Le in 2019, and it combines a neural-architecture-search baseline (EfficientNet-B0) with a scaling rule applied to produce B1 through B7.[1](https://doi.org/10.48550/arxiv.1905.11946) • [2](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/) EfficientNet-B0 matches ResNet-50's ImageNet accuracy with 5.3M parameters and 0.39B FLOPS, against ResNet-50's 26M parameters and 4.1B FLOPS.[1](https://doi.org/10.48550/arxiv.1905.11946)\n\n| Key fact | Value |\n|---|---|\n| Compound scaling rule | \\( d = \\alpha^{\\phi} \\), \\( w = \\beta^{\\phi} \\), \\( r = \\gamma^{\\phi} \\), with \\( \\alpha \\cdot \\beta^{2} \\cdot \\gamma^{2} \\approx 2 \\)[1](https://doi.org/10.48550/arxiv.1905.11946) |\n| B0 coefficients | \\( \\alpha = 1.2 \\), \\( \\beta = 1.1 \\), \\( \\gamma = 1.15 \\)[1](https://doi.org/10.48550/arxiv.1905.11946) |\n| B0 accuracy and size | 77.1% top-1 / 93.3% top-5 ImageNet, 5.3M parameters, 0.39B FLOPS[1](https://doi.org/10.48550/arxiv.1905.11946) |\n| B7 accuracy and size | 84.4% top-1 / 97.1% top-5 (84.3% in later tables), 66M parameters, 37B FLOPS[1](https://doi.org/10.48550/arxiv.1905.11946) • [3](https://www.nature.com/articles/s41598-025-04479-2/tables/1) |\n| Gain over single-axis scaling | +1.4% ImageNet accuracy on MobileNet, +0.7% on ResNet versus conventional scaling[2](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/) |\n| EfficientNetV2-M vs B7 | Comparable accuracy (85.1% vs 85.0% top-1), 17% fewer parameters, 37% fewer FLOPs, about 4.2x faster training[4](https://doi.org/10.48550/arxiv.2104.00298) |\n| GPU computational efficiency | 5–8% of ideal performance on an NVIDIA A5000, despite leading accuracy per MAC[5](https://arxiv.org/pdf/2404.03617) |\n\n## How it works\n\nCompound scaling answers a question earlier networks left open: given a compute budget, how much should a network grow in depth, width, and input resolution? MobileNets had offered separate single-dimension knobs, a width multiplier that reduces cost roughly by \\( \\alpha^{2} \\) and a resolution multiplier that reduces cost by \\( \\rho^{2} \\),[6](https://doi.org/10.48550/arxiv.1704.04861) but practitioners had to tune them independently. EfficientNet ties all three to one coefficient \\( \\phi \\):[1](https://doi.org/10.48550/arxiv.1905.11946)\n\n\\[ d = \\alpha^{\\phi}, \\quad w = \\beta^{\\phi}, \\quad r = \\gamma^{\\phi}, \\quad \\text{s.t.} \\quad \\alpha \\cdot \\beta^{2} \\cdot \\gamma^{2} \\approx 2 \\]\n\nThe constraint makes total FLOPS grow approximately as \\( 2^{\\phi} \\), because doubling depth doubles FLOPS while doubling width or resolution raises FLOPS about fourfold.[1](https://doi.org/10.48550/arxiv.1905.11946) The coefficients come from a small grid search on the baseline network under a fixed resource constraint, such as 2x more FLOPS; for B0 the search found \\( \\alpha = 1.2 \\), \\( \\beta = 1.1 \\), \\( \\gamma = 1.15 \\), and fixing them while varying \\( \\phi \\) yields B1 through B7.[1](https://doi.org/10.48550/arxiv.1905.11946) • [2](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/)\n\nThe rule measurably beats single-axis scaling. Applied to ResNet-50, compound scaling raises top-1 accuracy from 76.0% (4.1B FLOPS) to 78.8% at 16.7B FLOPS, versus 78.1% for depth-only scaling at 16.2B FLOPS; on MobileNetV1 it reaches 75.6% at 2.3B FLOPS versus 74.2% for width-only scaling at 2.2B FLOPS.[1](https://doi.org/10.48550/arxiv.1905.11946)\n\n## How it is done\n\nBuilding an EfficientNet has two stages. First, the B0 baseline is found by multi-objective neural architecture search in the MnasNet search space, optimizing \\( \\mathrm{ACC}(m) \\times [\\mathrm{FLOPS}(m)/T]^{w} \\) with \\( w = -0.07 \\) and a 400M FLOPS target.[1](https://doi.org/10.48550/arxiv.1905.11946) The main building block is the mobile inverted bottleneck (MBConv), an inverted residual block in which 1x1 convolutions expand the channel dimension before the 3x3 convolution, with squeeze-and-excitation channel attention added.[1](https://doi.org/10.48550/arxiv.1905.11946)\n\nSecond, the grid-searched coefficients are applied at increasing \\( \\phi \\) to produce B1–B7. Training uses [RMSProp](https://www.edgepedia.org/rmsprop) with decay 0.9 and momentum 0.9, batch norm momentum 0.99, weight decay 1e-5, an initial learning rate of 0.256 decaying by 0.97 every 2.4 epochs, AutoAugment, stochastic depth with survival probability 0.8, and dropout increased linearly from 0.2 (B0) to 0.5 (B7).[1](https://doi.org/10.48550/arxiv.1905.11946) Keras and TorchVision both ship the family; Keras applies width and depth coefficients internally (B7: 2.0/3.1 at 600px, dropout 0.5) and includes input rescaling in the model, while TorchVision exposes B0–B7 plus EfficientNetV2 S/M/L with pretrained weights.[7](https://github.com/tensorflow/tensorflow/blob/v2.5.0/tensorflow/python/keras/applications/efficientnet.py) • [8](https://docs.pytorch.org/vision/main/models/efficientnet.html)\n\n## Origin\n\nEfficientNet assembles several earlier components. MnasNet contributed the latency-aware search space and the AutoML MNAS framework used to find B0.[9](https://arxiv.org/abs/1807.11626) • [2](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/) The squeeze-and-excitation block, which adaptively recalibrates channel-wise feature responses by modeling interdependencies between channels, comes from Hu, Shen, Albanie, Sun, and Wu, whose SE networks won ILSVRC 2017 classification with 2.251% top-5 error.[10](https://doi.org/10.1109/tpami.2019.2913372) The MBConv block itself was originally developed for MobileNetV2 as a memory-efficient block for small mobile devices, and MobileNets' depthwise separable convolutions and width/resolution multipliers supplied the single-dimension scaling knobs that compound scaling unified.[5](https://arxiv.org/pdf/2404.03617) • [6](https://doi.org/10.48550/arxiv.1704.04861) [Stochastic](https://www.edgepedia.org/stochastic) depth, used in training, is an earlier regularization technique from Huang, Sun, Liu, Sedra, and Weinberger.[11](https://doi.org/10.48550/arxiv.1603.09382) The B7 result was reported against [GPipe](https://www.edgepedia.org/gpipe), a pipeline-parallelism approach that had set the previous accuracy benchmark.[1](https://doi.org/10.48550/arxiv.1905.11946) • [12](https://doi.org/10.48550/arxiv.1811.06965)\n\n## Variants\n\nThe core ladder runs B0 through B7. A 2025 peer-reviewed table lists top-1 accuracies of 77.1%, 79.1%, 80.1%, 81.6%, 82.9%, 83.6%, 84.0%, and 84.3% for B0–B7, with parameters from 5.3M to 66M and FLOPs from 0.39B to 37.0B.[3](https://www.nature.com/articles/s41598-025-04479-2/tables/1) The original paper prints 84.4% top-1 for B7.[1](https://doi.org/10.48550/arxiv.1905.11946)\n\nPost-paper recipes raised accuracy further: AutoAugment gives B7 84.3%, RandAugment 84.7%, AdvProp+AA 85.2% (B8 85.5%), and NoisyStudent+RA 86.9% top-1, with L2-475 and EfficientNet-L2 reaching 88.2% and 88.4% using extra JFT-300M unlabeled data.[13](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/README.md) EfficientNet-Lite (2020) removes squeeze-and-excitation, which some mobile accelerators support poorly, and replaces swish with RELU6 for easier post-quantization; lite0 has 4.7M parameters, 75.1% FP32 accuracy, and 74.4% INT8 accuracy, with FP32 and INT8 TFLite files shipped per checkpoint.[14](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/lite/README.md)\n\nEfficientNetV2, reported by Tan and Le in 2021, uses training-aware NAS over a search space enriched with Fused-MBConv ops, which replace MBConv's depthwise conv3x3 and expansion conv1x1 with a single regular conv3x3, and progressive learning, which adapts regularization to image size across four training stages of about 87 epochs each.[4](https://doi.org/10.48550/arxiv.2104.00298) EfficientNetV2-M matches B7's accuracy while training about 4.2x faster (13h vs 54h) with 17% fewer parameters and 37% fewer FLOPs.[4](https://doi.org/10.48550/arxiv.2104.00298)\n\n## Applications\n\nEfficientNets transfer well: they achieved state-of-the-art accuracy on 5 of 8 transfer datasets, including CIFAR-100 (91.7%) and Flowers (98.8%), with up to 21x fewer parameters.[2](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/) A 2024 benchmark of lightweight backbones recommends EfficientNet and RegNet for fine-tuning across diverse domains including remote sensing, plant, and medical (histopathology) datasets, and notes that for low-data regimes pure CNN backbones such as EfficientNet outperform transformers like Swin.[15](https://arxiv.org/pdf/2406.05612) EfficientNet-Lite serves edge deployment through INT8 TFLite files.[14](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/lite/README.md) Against the pre-2019 state of the art, B7's 84.4% top-1 came with 8.4x fewer parameters and 6.1x faster CPU inference than GPipe (3.1s vs 19.0s), and B1 runs 5.7x faster than ResNet-152 at higher accuracy (78.8% at 0.098s vs 77.8% at 0.554s).[1](https://doi.org/10.48550/arxiv.1905.11946) EfficientNetV2-L reaches 85.7% top-1, surpassing ViT-L/16(21k), and with ImageNet-21k pretraining the family hits 87.3% top-1, 2.0% above the compared ViT while training 5x–11x faster.[4](https://doi.org/10.48550/arxiv.2104.00298)\n\n## Limitations and alternatives\n\nLow FLOPs do not mean low latency. Measured on an NVIDIA A5000 GPU with PyTorch Inductor, EfficientNet's computational efficiency, the ratio of actual to ideal performance, is only 5% to 8%, and in practice it ran slower than many older models on GPUs with available inference engines, because depthwise convolutions have low operational intensity and few early-stage channels.[5](https://arxiv.org/pdf/2404.03617) The EfficientNet-X rework is more than 2x faster than EfficientNet on TPUv3 and GPUv100 at comparable accuracy, and latency-aware compound scaling (LACS) finds that depth should grow much faster than image size and width, unlike the original accuracy-only scaling, adding 14% and 25% average speedup on TPUs and GPUs respectively.[16](https://ar5iv.labs.arxiv.org/html/2102.05610) Google's own follow-up lists three training bottlenecks: very large image sizes slow accelerators through memory use, depthwise convolutions use hardware poorly, and uniform scaling of every stage is sub-optimal.[4](https://doi.org/10.48550/arxiv.2104.00298)\n\nDeployment and transfer carry further caveats. EfficientNet's specialized operations (depthwise separable convolutions, squeeze-and-excitation) are often unsupported or inefficient on MCU/NPU hardware, and the model is frequently excluded from ultra-low-bit quantization benchmarks because its layers typically require quantization-aware training to recover accuracy.[17](https://arxiv.org/pdf/2601.05364) High pretraining accuracy does not predict fine-tuning results: EfficientNetV2-S has the highest pretraining ImageNet-1k accuracy among tested backbones (84.23%) yet ranked best in none of the evaluated domains.[15](https://arxiv.org/pdf/2406.05612)\n\nNewer architectures are the main alternatives. ConvNeXt, a modernized pure-ConvNet family, competes favorably with EfficientNet in accuracy-computation trade-off and outperforms EfficientNetV2 when both use ImageNet-22K pretraining, reaching 87.8% top-1 at ConvNeXt-XL.[18](https://arxiv.org/pdf/2201.03545) In a controlled 2025-2026 benchmark, EfficientNetV2-S recorded the highest top-1 on CIFAR-10 (97.57%) and CIFAR-100 (86.98%), and RepViT-M1.0 led Tiny ImageNet (79.87%), while B0 sat on every bivariate accuracy-resource Pareto frontier across six resource dimensions despite using about 79% fewer parameters and 86% fewer GMACs than V2-S.[19](https://arxiv.org/pdf/2607.01984.pdf) Head-to-head numbers against DenseNet, MobileNetV3, and RegNet at matched compute, and post-2023 ONNX adoption specifics, are not covered by published comparisons.\n\n## References\n\n1. [Tan, Mingxing, Le, Quoc V. (2019). EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1905.11946)\n2. [EfficientNet: Improving Accuracy and Efficiency through AutoML and Model Scaling](https://research.google/blog/efficientnet-improving-accuracy-and-efficiency-through-automl-and-model-scaling/)\n3. [Table 1: Performance results on ImageNet for EfficientNet variants (Scientific Reports, 2025)](https://www.nature.com/articles/s41598-025-04479-2/tables/1)\n4. [Tan, Mingxing, Le, Quoc V. (2021). EfficientNetV2: Smaller Models and Faster Training. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.2104.00298)\n5. [ConvFirst: A hardware-aware analysis of convolutional block efficiency (EfficientNet computational-efficiency critique)](https://arxiv.org/pdf/2404.03617)\n6. [Howard, Andrew G. and colleagues (2017). MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1704.04861)\n7. [Keras Applications EfficientNet source (TensorFlow v2.5.0)](https://github.com/tensorflow/tensorflow/blob/v2.5.0/tensorflow/python/keras/applications/efficientnet.py)\n8. [EfficientNet, TorchVision main documentation](https://docs.pytorch.org/vision/main/models/efficientnet.html)\n9. [MnasNet: Platform-Aware Neural Architecture Search for Mobile](https://arxiv.org/abs/1807.11626)\n10. [Jie Hu and colleagues (2019). Squeeze-and-Excitation Networks. IEEE Transactions on Pattern Analysis and Machine Intelligence.](https://doi.org/10.1109/tpami.2019.2913372)\n11. [Huang, Gao and colleagues (2016). Deep Networks with Stochastic Depth. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1603.09382)\n12. [Huang, Yanping and colleagues (2018). GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism. arXiv (Cornell University).](https://doi.org/10.48550/arxiv.1811.06965)\n13. [Official EfficientNet README (tensorflow/tpu)](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/README.md)\n14. [EfficientNet-Lite README (tensorflow/tpu)](https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/lite/README.md)\n15. [Benchmark study of lightweight CNN backbones for fine-tuning across image domains](https://arxiv.org/pdf/2406.05612)\n16. [Searching for Fast Model Families on Datacenter Accelerators (EfficientNet-X)](https://ar5iv.labs.arxiv.org/html/2102.05610)\n17. [CompressNAS / STResNet: ultra-compact models for MCU and NPU deployment](https://arxiv.org/pdf/2601.05364)\n18. [A ConvNet for the 2020s (ConvNeXt)](https://arxiv.org/pdf/2201.03545)\n19. [Do Newer Lightweight CNNs Perform Better Under Resource Constraints? A Controlled Multigenerational Study](https://arxiv.org/pdf/2607.01984.pdf)\n\n---\n*Topic: Encyclopedia › Technology and the built world › Computing and digital systems › Artificial intelligence and data › Machine learning and neural computation › Neural networks and deep learning › Titles A to H*\n\n*Initially written Sep 29, 2026 · Reviewed: Sep 30, 2026 · Edited: Sep 30, 2026 · Last review: Sep 30, 2026*\n\n*Copyright 2026 EdgeChat AI, a subsidiary of Biostate AI.*\n\nLicense: Edgepedia Community License 1.0, https://www.edgepedia.org/edgepedia/license\n", "same_as": [], "url": "https://www.edgepedia.org/efficientnet", "markdown_url": "https://www.edgepedia.org/efficientnet.md", "license": { "name": "Edgepedia Community License 1.0", "url": "https://www.edgepedia.org/edgepedia/license", "summary": "Free with credit, commercial use included. AI training is open to everyone. For other uses, organizations over USD 100M in revenue or 100M monthly users license separately.", "spdx": "LicenseRef-Edgepedia-Community-1.0" }, "credit": "\"EfficientNet\", Edgepedia (EdgeChat), https://www.edgepedia.org/efficientnet. Edgepedia Community License 1.0.", "credit_md": "\"[EfficientNet](https://www.edgepedia.org/efficientnet)\", Edgepedia (EdgeChat), [https://www.edgepedia.org/efficientnet](https://www.edgepedia.org/efficientnet). [Edgepedia Community License 1.0](https://www.edgepedia.org/edgepedia/license).", "credit_html": "\"EfficientNet\", Edgepedia (EdgeChat), https://www.edgepedia.org/efficientnet. Edgepedia Community License 1.0.", "speakable": "EfficientNet is a family of convolutional neural networks for image classification that scales depth, width, and resolution together, reported by Mingxing Tan and Quoc V. Le in 2019." }