# ViCToR: Improving Visual Comprehension via Token Reconstruction for Pretraining LMMs

arXiv: https://arxiv.org/abs/2410.14332 · HTML: https://mlsanity.com/paper/2410.14332/

## Key facts

- Cited by 2 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 21 of its 34 references that are in the map.
- First posted on arXiv 18 Oct 2024.

## Placed in

- Multimodal LLMs / vision-language models / [Visual token pruning and efficiency](https://mlsanity.com/mllm/visual-token-pruning-and-efficiency/)

## Builds on (most cited first, 21 of 21)

- [Learning Transferable Visual Models From Natural Language Supervision](https://mlsanity.com/paper/2103.00020/) (2021)
- [LLaMA: Open and Efficient Foundation Language Models](https://mlsanity.com/paper/2302.13971/) (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)
- [Taming Transformers for High-Resolution Image Synthesis](https://mlsanity.com/paper/2012.09841/) (2020)
- [LLaVA-OneVision: Easy Visual Task Transfer](https://mlsanity.com/paper/2408.03326/) (2024)
- [MMBench: Is Your Multi-modal Model an All-around Player?](https://mlsanity.com/paper/2307.06281/) (2023)
- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Conceptual 12M: Pushing Web-Scale Image-Text Pre-Training To Recognize Long-Tail Visual Concepts](https://mlsanity.com/paper/2102.08981/) (2021)
- [ShareGPT4V: Improving Large Multi-Modal Models with Better Captions](https://mlsanity.com/paper/2311.12793/) (2023)
- [SEED-Bench: Benchmarking Multimodal LLMs with Generative Comprehension](https://mlsanity.com/paper/2307.16125/) (2023)
- [Cambrian-1: A Fully Open, Vision-Centric Exploration of Multimodal LLMs](https://mlsanity.com/paper/2406.16860/) (2024)
- [A Diagram Is Worth A Dozen Images](https://mlsanity.com/paper/1603.07396/) (2016)
- [VILA: On Pre-training for Visual Language Models](https://mlsanity.com/paper/2312.07533/) (2023)
- [EVA: Exploring the Limits of Masked Visual Representation Learning at Scale](https://mlsanity.com/paper/2211.07636/) (2022)
- [VLMEvalKit: An Open-Source Toolkit for Evaluating Large Multi-Modality Models](https://mlsanity.com/paper/2407.11691/) (2024)
- [MANTIS: Interleaved Multi-Image Instruction Tuning](https://mlsanity.com/paper/2405.01483/) (2024)
- [Unmasked Teacher: Towards Training-Efficient Video Foundation Models](https://mlsanity.com/paper/2303.16058/) (2023)
- [MVP: Multimodality-guided Visual Pre-training](https://mlsanity.com/paper/2203.05175/) (2022)
- [MILAN: Masked Image Pretraining on Language Assisted Representation](https://mlsanity.com/paper/2208.06049/) (2022)
- [RILS: Masked Visual Reconstruction in Language Semantic Space](https://mlsanity.com/paper/2301.06958/) (2023)

## Built on by (most cited first, 2 of 2)

- [From Illusion to Intention: Visual Rationale Learning for Vision-Language Reasoning](https://mlsanity.com/paper/2511.23031/) (2025)
- [ORID: Organ-Regional Information Driven Framework for Radiology Report Generation](https://mlsanity.com/paper/2411.13025/) (2024)

## Abstract (arXiv)

Large Multimodal Models (LMMs) often face a modality representation gap during pretraining: while language embeddings remain stable, visual representations are highly sensitive to contextual noise (e.g., background clutter). To address this issue, we introduce a visual comprehension stage, which we call ViCToR (Visual Comprehension via Token Reconstruction), a novel pretraining framework for LMMs. ViCToR employs a learnable visual token pool and utilizes the Hungarian matching algorithm to select semantically relevant tokens from this pool for visual token replacement. Furthermore, by integrating a visual token reconstruction loss with dense semantic supervision, ViCToR can learn tokens which retain high visual detail, thereby enhancing the large language model's (LLM's) understanding of visual information. After pretraining on 3 million publicly accessible images and captions, ViCToR achieves state-of-the-art results, improving over LLaVA-NeXT-8B by 10.4%, 3.2%, and 7.2% on the MMStar, SEEDI, and RealWorldQA benchmarks, respectively. Code is available at https://github.com/deepglint/Victor.