# EMOv2: Pushing 5M Vision Model Frontier

arXiv: https://arxiv.org/abs/2412.06674 · HTML: https://mlsanity.com/paper/2412.06674/

## Key facts

- Cited by 5 papers in the map (data through 14 Oct 2026).
- Builds on 31 of its 115 references that are in the map.
- First posted on arXiv 9 Dec 2024.

## Builds on (most cited first, 30 of 31)

- [Layer Normalization](https://mlsanity.com/paper/1607.06450/) (2016)
- [Mamba: Linear-Time Sequence Modeling with Selective State Spaces](https://mlsanity.com/paper/2312.00752/) (2023)
- [MobileNets: Efficient Convolutional Neural Networks for Mobile Vision Applications](https://mlsanity.com/paper/1704.04861/) (2017)
- [Gaussian Error Linear Units (GELUs)](https://mlsanity.com/paper/1606.08415/) (2016)
- [Rethinking Atrous Convolution for Semantic Image Segmentation](https://mlsanity.com/paper/1706.05587/) (2017)
- [The Kinetics Human Action Video Dataset](https://mlsanity.com/paper/1705.06950/) (2017)
- [MMDetection: Open MMLab Detection Toolbox and Benchmark](https://mlsanity.com/paper/1906.07155/) (2019)
- [SqueezeNet: AlexNet-level accuracy with 50x fewer parameters and <0.5MB model size](https://mlsanity.com/paper/1602.07360/) (2016)
- [Vision Mamba: Efficient Visual Representation Learning with Bidirectional State Space Model](https://mlsanity.com/paper/2401.09417/) (2024)
- [RWKV: Reinventing RNNs for the Transformer Era](https://mlsanity.com/paper/2305.13048/) (2023)
- [SiT: Exploring Flow and Diffusion-based Generative Models with Scalable Interpolant Transformers](https://mlsanity.com/paper/2401.08740/) (2024)
- [EfficientVMamba: Atrous Selective Scan for Light Weight Visual Mamba](https://mlsanity.com/paper/2403.09977/) (2024)
- [MambaOut: Do We Really Need Mamba for Vision?](https://mlsanity.com/paper/2405.07992/) (2024)
- [RepViT: Revisiting Mobile CNN From ViT Perspective](https://mlsanity.com/paper/2307.09283/) (2023)
- [MobileNetV4 -- Universal Models for the Mobile Ecosystem](https://mlsanity.com/paper/2404.10518/) (2024)
- [Vision-RWKV: Efficient and Scalable Visual Perception with RWKV-Like Architectures](https://mlsanity.com/paper/2403.02308/) (2024)
- [Next-ViT: Next Generation Vision Transformer for Efficient Deployment in Realistic Industrial Scenarios](https://mlsanity.com/paper/2207.05501/) (2022)
- [Multi-Scale VMamba: Hierarchy in Hierarchy Visual State Space Model](https://mlsanity.com/paper/2405.14174/) (2024)
- [MobileViTv3: Mobile-Friendly Vision Transformer with Simple and Effective Fusion of Local, Global and Input Features](https://mlsanity.com/paper/2209.15159/) (2022)
- [EdgeSAM: Prompt-In-the-Loop Distillation for SAM](https://mlsanity.com/paper/2312.06660/) (2023)
- [LightViT: Towards Light-Weight Convolution-Free Vision Transformers](https://mlsanity.com/paper/2207.05557/) (2022)
- [Visual Attention Methods in Deep Learning: An In-Depth Survey](https://mlsanity.com/paper/2204.07756/) (2022)
- [TinySAM: Pushing the Envelope for Efficient Segment Anything Model](https://mlsanity.com/paper/2312.13789/) (2023)
- [Recent Advances in Vision Transformer: A Survey and Outlook of Recent Work](https://mlsanity.com/paper/2203.01536/) (2022)
- [PointRWKV: Efficient RWKV-Like Model for Hierarchical Point Cloud Learning](https://mlsanity.com/paper/2405.15214/) (2024)
- [Mamba or RWKV: Exploring High-Quality and High-Efficiency Segment Anything Model](https://mlsanity.com/paper/2406.19369/) (2024)
- [MobileMamba: Lightweight Multi-Receptive Visual Mamba Network](https://mlsanity.com/paper/2411.15941/) (2024)
- [GhostNetV3: Exploring the Training Strategies for Compact Models](https://mlsanity.com/paper/2404.11202/) (2024)
- [ViG: Linear-complexity Visual Sequence Learning with Gated Linear Attention](https://mlsanity.com/paper/2405.18425/) (2024)
- [MoCoViT: Mobile Convolutional Vision Transformer](https://mlsanity.com/paper/2205.12635/) (2022)

## Built on by (most cited first, 5 of 5)

- [VQualA 2025 Challenge on Face Image Quality Assessment: Methods and Results](https://mlsanity.com/paper/2508.18445/) (2025)
- [S2AFormer: Strip Self-Attention for Efficient Vision Transformer](https://mlsanity.com/paper/2505.22195/) (2025)
- [Transform Trained Transformer for Accelerating Native 4K Video Generation](https://mlsanity.com/paper/2512.13492/) (2025)
- [PixVerve: Advancing Native UHR Image Generation to 100MP with a Large-Scale High-Quality Dataset](https://mlsanity.com/paper/2605.20147/) (2026)
- [PiT: Progressive Diffusion Transformer](https://mlsanity.com/paper/2505.13219/) (2025)

## Abstract (arXiv)

This work focuses on developing parameter-efficient and lightweight models for dense predictions while trading off parameters, FLOPs, and performance. Our goal is to set up the new frontier of the 5M magnitude lightweight model on various downstream tasks. Inverted Residual Block (IRB) serves as the infrastructure for lightweight CNNs, but no counterparts have been recognized by attention-based design. Our work rethinks the lightweight infrastructure of efficient IRB and practical components in Transformer from a unified perspective, extending CNN-based IRB to attention-based models and abstracting a one-residual Meta Mobile Block (MMBlock) for lightweight model design. Following neat but effective design criterion, we deduce a modern Improved Inverted Residual Mobile Block (i2RMB) and improve a hierarchical Efficient MOdel (EMOv2) with no elaborate complex structures. Considering the imperceptible latency for mobile users when downloading models under 4G/5G bandwidth and ensuring model performance, we investigate the performance upper limit of lightweight models with a magnitude of 5M. Extensive experiments on various vision recognition, dense prediction, and image generation tasks demonstrate the superiority of our EMOv2 over state-of-the-art methods, e.g., EMOv2-1M/2M/5M achieve 72.3, 75.8, and 79.4 Top-1 that surpass equal-order CNN-/Attention-based models significantly. At the same time, EMOv2-5M equipped RetinaNet achieves 41.5 mAP for object detection tasks that surpasses the previous EMO-5M by +2.6. When employing the more robust training recipe, our EMOv2-5M eventually achieves 82.9 Top-1 accuracy, which elevates the performance of 5M magnitude models to a new level. Code is available at https://github.com/zhangzjn/EMOv2.