# HiDrop: Hierarchical Vision Token Reduction in MLLMs via Late Injection, Concave Pyramid Pruning, and Early Exit

arXiv: https://arxiv.org/abs/2602.23699 · HTML: https://mlsanity.com/paper/2602.23699/

## Key facts

- Cited by 18 papers in the map (data through 14 Oct 2026).
- Placed in Visual token pruning and efficiency, one of 16 directions in Multimodal LLMs / vision-language models.
- Builds on 40 of its 48 references that are in the map.
- First posted on arXiv 27 Feb 2026.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)
- Efficient LLM inference and serving / [Visual token pruning in multimodal models](https://mlsanity.com/inference/visual-token-pruning-in-multimodal-models/)

## Builds on (most cited first, 30 of 40)

- [An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale](https://mlsanity.com/paper/2010.11929/) (2020)
- [GPT-4 Technical Report](https://mlsanity.com/paper/2303.08774/) (2023)
- [Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena](https://mlsanity.com/paper/2306.05685/) (2023)
- [BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models](https://mlsanity.com/paper/2301.12597/) (2023)
- [Qwen2.5-VL Technical Report](https://mlsanity.com/paper/2502.13923/) (2025)
- [Improved Baselines with Visual Instruction Tuning](https://mlsanity.com/paper/2310.03744/) (2023)
- [Qwen2-VL: Enhancing Vision-Language Model's Perception of the World at Any Resolution](https://mlsanity.com/paper/2409.12191/) (2024)
- [Making the V in VQA Matter: Elevating the Role of Image Understanding in Visual Question Answering](https://mlsanity.com/paper/1612.00837/) (2016)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering](https://mlsanity.com/paper/2209.09513/) (2022)
- [Towards VQA Models That Can Read](https://mlsanity.com/paper/1904.08920/) (2019)
- [MME: A Comprehensive Evaluation Benchmark for Multimodal Large Language Models](https://mlsanity.com/paper/2306.13394/) (2023)
- [Are We on the Right Way for Evaluating Large Vision-Language Models?](https://mlsanity.com/paper/2403.20330/) (2024)
- [VizWiz Grand Challenge: Answering Visual Questions from Blind People](https://mlsanity.com/paper/1802.08218/) (2018)
- [DeepSeek-VL: Towards Real-World Vision-Language Understanding](https://mlsanity.com/paper/2403.05525/) (2024)
- [An Image is Worth 1/2 Tokens After Layer 2: Plug-and-Play Inference Acceleration for Large Vision-Language Models](https://mlsanity.com/paper/2403.06764/) (2024)
- [SparseVLM: Visual Token Sparsification for Efficient Vision-Language Model Inference](https://mlsanity.com/paper/2410.04417/) (2024)
- [LMMs-Eval: Reality Check on the Evaluation of Large Multimodal Models](https://mlsanity.com/paper/2407.12772/) (2024)
- [VisionZip: Longer is Better but Not Necessary in Vision Language Models](https://mlsanity.com/paper/2412.04467/) (2024)
- [PyramidDrop: Accelerating Your Large Vision-Language Models via Pyramid Visual Redundancy Reduction](https://mlsanity.com/paper/2410.17247/) (2024)
- [A Survey on Multimodal Large Language Models for Autonomous Driving](https://mlsanity.com/paper/2311.12320/) (2023)
- [Honeybee: Locality-enhanced Projector for Multimodal LLM](https://mlsanity.com/paper/2312.06742/) (2023)
- [TokenPacker: Efficient Visual Projector for Multimodal LLM](https://mlsanity.com/paper/2407.02392/) (2024)
- [LLaVA-Mini: Efficient Image and Video Large Multimodal Models with One Vision Token](https://mlsanity.com/paper/2501.03895/) (2025)
- [ATP-LLaVA: Adaptive Token Pruning for Large Vision Language Models](https://mlsanity.com/paper/2412.00447/) (2024)
- [VoCo-LLaMA: Towards Vision Compression with Large Language Models](https://mlsanity.com/paper/2406.12275/) (2024)
- [Multi-Stage Vision Token Dropping: Towards Efficient Multimodal Large Language Model](https://mlsanity.com/paper/2411.10803/) (2024)
- [Dynamic-LLaVA: Efficient Multimodal Large Language Models via Dynamic Vision-language Context Sparsification](https://mlsanity.com/paper/2412.00876/) (2024)
- [MobileVLM: A Vision-Language Model for Better Intra- and Inter-UI Understanding](https://mlsanity.com/paper/2409.14818/) (2024)
- [Efficient Large Multi-modal Models via Visual Context Compression](https://mlsanity.com/paper/2406.20092/) (2024)

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

- [Think-as-You-See: Streaming Chain-of-Thought Reasoning for Large Vision-Language Models](https://mlsanity.com/paper/2603.02872/) (2026)
- [Beyond FLOPs: Benchmarking Real Inference Acceleration of LLM Pruning under a GEMM-Centric Taxonomy](https://mlsanity.com/paper/2606.09080/) (2026)
- [From Static Inference to Dynamic Interaction: A Survey of Streaming Large Language Models](https://mlsanity.com/paper/2603.04592/) (2026)
- [SinkPruner: Sink-Free Visual Token Pruning for Multimodal Large Language Models](https://mlsanity.com/paper/2609.01004/) (2026)
- [PathSelect: Sequential Token Selection for Whole Slide Pathology](https://mlsanity.com/paper/2607.23631/) (2026)
- [ViCoStream: Streaming VideoLLMs Can Run Beyond 100 FPS with Stage-Wise Coordinated Inference](https://mlsanity.com/paper/2606.19849/) (2026)
- [WIDE: Boosting Adaptive LLM Inference via Token-level Dynamic Width Pruning](https://mlsanity.com/paper/2607.28418/) (2026)
- [VideoRouter: Query-Adaptive Dual Routing for Efficient Long-Video Understanding](https://mlsanity.com/paper/2605.05848/) (2026)
- [Attention-Free and Lightweight Token Reduction for Efficient Vision-Language Models](https://mlsanity.com/paper/2607.13500/) (2026)
- [Beyond Single-Dimensional Compression: The Compound Sparsity Frontier of Large Language Models](https://mlsanity.com/paper/2607.18280/) (2026)
- [From Saliency to Discriminability: Rank-Preserving Visual Token Pruning for VLM Rerankers](https://mlsanity.com/paper/2609.00667/) (2026)
- [Shallow to Deep: Aligning Token Pruning with Stage-wise Roles in LVLMs](https://mlsanity.com/paper/2609.25635/) (2026)
- [CoViST: Visual Token Compression via Composable States](https://mlsanity.com/paper/2609.33397/) (2026)
- [ACPruner: Visual Token Pruning as Biased Attention Coverage Maximization in LVLMs](https://mlsanity.com/paper/2609.34558/) (2026)
- [V-CoLA: Vision Token Compression with Linear Attention](https://mlsanity.com/paper/2610.11251/) (2026)
- [GeoGR: Spatial-Semantic Token Compression and Group Routing for Efficient 3D VLMs](https://mlsanity.com/paper/2610.13035/) (2026)
- [Bifocal: Training-Free Mixed Granularity Visual Token Pruning for Multimodal LLMs](https://mlsanity.com/paper/2610.18399/) (2026)
- [Just Respect the Semantics: Image-Token Compression via Pre-LLM Semantic Variable-Resolution Tokenization and In-LLM Relevance Pruning](https://mlsanity.com/paper/2610.18690/) (2026)

## Abstract (arXiv)

The quadratic computational cost of processing vision tokens in Multimodal Large Language Models (MLLMs) hinders their widespread adoption. While progressive vision token pruning offers a promising solution, current methods misinterpret shallow layer functions and use rigid schedules, which fail to unlock the full efficiency potential. To address these issues, we propose HiDrop, a framework that aligns token pruning with the true hierarchical function of MLLM layers. HiDrop features two key innovations: (1) Late Injection, which bypasses passive shallow layers to introduce visual tokens exactly where active fusion begins; and (2) Concave Pyramid Pruning with an Early Exit mechanism to dynamically adjust pruning rates across middle and deep layers. This process is optimized via an inter-layer similarity measure and a differentiable top-k operator. To ensure practical efficiency, HiDrop further incorporates persistent positional encoding, FlashAttention-compatible token selection, and parallel decoupling of vision computation to eliminate hidden overhead associated with dynamic token reduction. Extensive experiments show that HiDrop compresses about 90% visual tokens while matching the original performance and accelerating training by 1.72 times. Our work not only sets a new state-of-the-art for efficient MLLM training and inference but also provides valuable insights into the hierarchical nature of multimodal fusion. The code is released at https://github.com/EIT-NLP/HiDrop.