# Window Attention is Bugged: How not to Interpolate Position Embeddings

arXiv: https://arxiv.org/abs/2311.05613 · HTML: https://mlsanity.com/paper/2311.05613/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- Builds on 48 of its 52 references that are in the map.
- First posted on arXiv 9 Nov 2023.

## Builds on (most cited first, 30 of 48)

- [Attention Is All You Need](https://mlsanity.com/paper/1706.03762/) (2017)
- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [Microsoft COCO: Common Objects in Context](https://mlsanity.com/paper/1405.0312/) (2014)
- [Swin Transformer: Hierarchical Vision Transformer using Shifted Windows](https://mlsanity.com/paper/2103.14030/) (2021)
- [Segment Anything](https://mlsanity.com/paper/2304.02643/) (2023)
- [End-to-End Object Detection with Transformers](https://mlsanity.com/paper/2005.12872/) (2020)
- [Masked Autoencoders Are Scalable Vision Learners](https://mlsanity.com/paper/2111.06377/) (2021)
- [Mask R-CNN](https://mlsanity.com/paper/1703.06870/) (2017)
- [EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks](https://mlsanity.com/paper/1905.11946/) (2019)
- [Rethinking the Inception Architecture for Computer Vision](https://mlsanity.com/paper/1512.00567/) (2015)
- [mixup: Beyond Empirical Risk Minimization](https://mlsanity.com/paper/1710.09412/) (2017)
- [Feature Pyramid Networks for Object Detection](https://mlsanity.com/paper/1612.03144/) (2016)
- [RoFormer: Enhanced Transformer with Rotary Position Embedding](https://mlsanity.com/paper/2104.09864/) (2021)
- [FlashAttention: Fast and Memory-Efficient Exact Attention with IO-Awareness](https://mlsanity.com/paper/2205.14135/) (2022)
- [CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features](https://mlsanity.com/paper/1905.04899/) (2019)
- [FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioning](https://mlsanity.com/paper/2307.08691/) (2023)
- [BEiT: BERT Pre-Training of Image Transformers](https://mlsanity.com/paper/2106.08254/) (2021)
- [RandAugment: Practical automated data augmentation with a reduced search space](https://mlsanity.com/paper/1909.13719/) (2019)
- [Pyramid Vision Transformer: A Versatile Backbone for Dense Prediction without Convolutions](https://mlsanity.com/paper/2102.12122/) (2021)
- [Rethinking Attention with Performers](https://mlsanity.com/paper/2009.14794/) (2020)
- [Linformer: Self-Attention with Linear Complexity](https://mlsanity.com/paper/2006.04768/) (2020)
- [VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training](https://mlsanity.com/paper/2203.12602/) (2022)
- [LVIS: A Dataset for Large Vocabulary Instance Segmentation](https://mlsanity.com/paper/1908.03195/) (2019)
- [SimMIM: A Simple Framework for Masked Image Modeling](https://mlsanity.com/paper/2111.09886/) (2021)
- [Self-Attention with Relative Position Representations](https://mlsanity.com/paper/1803.02155/) (2018)
- [Scaling Vision Transformers](https://mlsanity.com/paper/2106.04560/) (2021)
- [Multiscale Vision Transformers](https://mlsanity.com/paper/2104.11227/) (2021)
- [PVT v2: Improved Baselines with Pyramid Vision Transformer](https://mlsanity.com/paper/2106.13797/) (2021)
- [Exploring Plain Vision Transformer Backbones for Object Detection](https://mlsanity.com/paper/2203.16527/) (2022)
- [Detecting Twenty-thousand Classes using Image-level Supervision](https://mlsanity.com/paper/2201.02605/) (2022)

## Built on by (most cited first, 18 of 18)

- [SAM 2: Segment Anything in Images and Videos](https://mlsanity.com/paper/2408.00714/) (2024)
- [Perception Encoder: The best visual embeddings are not at the output of the network](https://mlsanity.com/paper/2504.13181/) (2025)
- [AM-RADIO: Agglomerative Vision Foundation Model -- Reduce All Domains Into One](https://mlsanity.com/paper/2312.06709/) (2023)
- [When Do We Not Need Larger Vision Models?](https://mlsanity.com/paper/2403.13043/) (2024)
- [CorrCLIP: Reconstructing Patch Correlations in CLIP for Open-Vocabulary Semantic Segmentation](https://mlsanity.com/paper/2411.10086/) (2024)
- [OpenWorldSAM: Extending SAM2 for Universal Image Segmentation with Language Prompts](https://mlsanity.com/paper/2507.05427/) (2025)
- [MemorySAM: Memorize Modalities and Semantics with Segment Anything Model 2 for Multi-modal Semantic Segmentation](https://mlsanity.com/paper/2503.06700/) (2025)
- [Boltzmann Attention Sampling for Image Analysis with Small Objects](https://mlsanity.com/paper/2503.02841/) (2025)
- [From SAM to SAM 2: Exploring Improvements in Meta's Segment Anything Model](https://mlsanity.com/paper/2408.06305/) (2024)
- [Fine-Grained Spatially Varying Material Selection in Images](https://mlsanity.com/paper/2506.09023/) (2025)
- [SAM4D: Segment Anything in Camera and LiDAR Streams](https://mlsanity.com/paper/2506.21547/) (2025)
- [Segment Anything Across Shots: A Method and Benchmark](https://mlsanity.com/paper/2511.13715/) (2025)
- [Efficient-SAM2: Accelerating SAM2 with Object-Aware Visual Encoding and Memory Retrieval](https://mlsanity.com/paper/2602.08224/) (2026)
- [Adapting Vision-Language Model with Fine-grained Semantics for Open-Vocabulary Segmentation](https://mlsanity.com/paper/2409.16278/) (2024)
- [DAM-Net: Domain Adaptation Network with Micro-Labeled Fine-Tuning for Change Detection](https://mlsanity.com/paper/2504.13748/) (2025)
- [Automated urban waterlogging assessment and early warning through a mixture of foundation models](https://mlsanity.com/paper/2510.18425/) (2025)
- [Eye image segmentation using visual and concept prompts with Segment Anything Model 3 (SAM3)](https://mlsanity.com/paper/2603.17715/) (2026)
- [OV-Stitcher: A Global Context-Aware Framework for Training-Free Open-Vocabulary Semantic Segmentation](https://mlsanity.com/paper/2604.08110/) (2026)

## Abstract (arXiv)

Window attention, position embeddings, and high resolution finetuning are core concepts in the modern transformer era of computer vision. However, we find that naively combining these near ubiquitous components can have a detrimental effect on performance. The issue is simple: interpolating position embeddings while using window attention is wrong. We study two state-of-the-art methods that have these three components, namely Hiera and ViTDet, and find that both do indeed suffer from this bug. To fix it, we introduce a simple absolute window position embedding strategy, which solves the bug outright in Hiera and allows us to increase both speed and performance of the model in ViTDet. We finally combine the two to obtain HieraDet, which achieves 61.7 box mAP on COCO, making it state-of-the-art for models that only use ImageNet-1k pretraining. This all stems from what is essentially a 3 line bug fix, which we name "absolute win".