通用视觉自监督研究报
A Daily Digest of Visual Self-Supervised Learning
2026-06-01 图像表征 · VFM · JEPA · 视频预训练
Visual SSL / representation · P2 · 2026-06-01

MergeTok：它和通用视觉自监督的关系在于：统一连续 VAE 与离散 VQ 视觉 tokenizer，用 token merging 做语义桥

中高相关；详见方法、贡献和实验边界。

MergeTok: Unified Continuous and Discrete Visual Tokenization via Token Merging　arXiv　原文链接

编号2605.30904 优先级P2 类别Visual SSL / representation 会议arXiv 方法统一连续 VAE 与离散 VQ 视觉 tokenizer，用 token merging 做语义桥 来源arXiv / OpenReview

先说结论。它和通用视觉自监督的关系在于：统一连续 VAE 与离散 VQ 视觉 tokenizer，用 token merging 做语义桥。 中高相关；详见方法、贡献和实验边界。

Figureure 2 · : Overall Framework of MergeTok
Figureure 2 · : Overall Framework of MergeTokFigure 2: Overall Framework of MergeTok. We propose a dual-branch architecture that jointly optimizes continuous and discrete representations with shared encoder and decoder. (i) VAE Branch (Bottom) applies ToMe [1] to extract dense semantic tokens, which are aligned with a teacher model (also equipped with ToMe). The resulting source map is then employed to unmerge the groups back to the full lattice for reconstruction. (ii) VQ Branch (Top) inherits this source map to induce group-aware clustering, enforcing intra-group diversity and inter-group exclusivity constraints that stabilize the training of the discrete codebook.这张图概括 MergeTok 的整体方法流程。阅读时先看模块之间传递的训练信号，再看作者如何把目标拆成可优化的子问题。
(c) MergeTok Tokenizer (Ours) Figure 1: (a) Discrete VQ
(c) MergeTok Tokenizer (Ours) Figure 1: (a) Discrete VQ(c) MergeTok Tokenizer (Ours) Figure 1: (a) Discrete VQ. Features are quantized by nearest-neighbor codebook lookup, but codebook updates can be sparse. (b) Continuous VAE. Features are mapped to continuous Gaussian latents through reparameterization for stable reconstruction. (c) MergeTok. MergeTok adopts a dual-branch design to jointly optimize VAE and VQ tokenization. The VAE branch introduces online token merging to inject semantic structure into continuous latents, while the resulting token-similarity information is used to guide group-aware VQ quantization, improving discrete token learning.这张图/表用于判断 MergeTok 的实验收益来自哪里。重点看替换、消融或跨模型设置下趋势是否一致，而不是只看单个最高分。

核心问题

它和通用视觉自监督的关系在于：统一连续 VAE 与离散 VQ 视觉 tokenizer，用 token merging 做语义桥。

方法拆解

统一连续 VAE 与离散 VQ 视觉 tokenizer，用 token merging 做语义桥

主要贡献

中高相关；详见方法、贡献和实验边界。

实验看点

实验部分建议重点看两类证据：一是作者是否把方法收益和更强数据、更长训练、更大模型区分开；二是跨模型、跨数据或跨任务迁移是否还能保留同样趋势。

局限与读法

这篇论文的结论需要结合任务设置、训练数据规模和消融实验一起看；不要只凭单个指标判断它对通用视觉表征的价值。