When Vision Becomes Text: Visual Token Pruning via Cross-Modal Residual Guidance in VLMs

Abundant visual information strengthens vision-language model (VLM) perception, yet massive visual tokens raise inference costs. Existing visual token pruning methods rely on similarity-based guidance, which exploits pairwise text-vision and vision-vision token correlations for compression. However, such methods only capture local layer-level signals and overlook the whole inference process in VLM. In this paper, we revisit VLM inference and present a new efficient guidance scheme that complements similarity-based guidance. In particular, we identify a key observation: as LLM layers deepen, text tokens continuously aggregate visual information via self-attention and progressively absorb partial visual content into textual representations. To quantify this phenomenon, we propose Cross Modal Absorption (CMA) from a geometric representation perspective to measure how much visual information is absorbed by text, revealing that more visual tokens in deeper layers can be approximately explained by the text subspace. We accordingly propose Cross Modal Residual (CMR). It projects visual tokens onto the text subspace via Tikhonov regularized least squares and exploits reconstruction residuals to quantify visual information that cannot be explained by text. Finally, based on CMR, we present SIEVE, a training-free visual token compression method that combines CMR, text-attention relevance, and residual-space diversity to retain task-relevant and complementary tokens. Experiments on diverse VLM architectures verify the effectiveness of SIEVE. For instance, on LLaVA-NeXT-7B, SIEVE keeps only $11.1\%$ of visual tokens while preserving $97.5\%$ of the original average performance, achieving $3.62\times$ prefill speedup, $2.49\times$ end-to-end speedup, and a $6.02\times$ KV-cache reduction.

PIO-FVLM: Rethinking Training-Free Visual Token Reduction for VLM Acceleration from an Inference-Objective Perspective

Recently, reducing redundant visual tokens in vision-language models (VLMs) to accelerate VLM inference has emerged as a hot topic. However, most existing methods rely on heuristics constructed based on inter-visual-token similarity or cross-modal visual-text similarity, which gives rise to certain limitations in compression performance and practical deployment. In contrast, we propose PIO-FVLM from the perspective of inference objectives, which transforms visual token compression into preserving output result invariance and selects tokens primarily by their importance to this goal. Specially, vision tokens are reordered with the guidance of token-level gradient saliency generated by our designed layer-local proxy loss, a coarse constraint from the current layer to the final result. Then the most valuable vision tokens are selected following the non-maximum suppression (NMS) principle. The proposed PIO-FVLM is training-free and compatible with FlashAttention, friendly to practical application and deployment. It can be deployed independently as an encoder-free method, or combined with encoder compression approaches like VisionZip for use as an encoder-involved method. On LLaVA-Next-7B, PIO-FVLM retains just 11.1% of visual tokens but maintains 97.2% of the original performance, with a 2.67$\times$ prefill speedup, 2.11$\times$ inference speedup, 6.22$\times$ lower FLOPs, and 6.05$\times$ reduced KV Cache overhead. Our code is available at https://github.com/ocy1/PIO-FVLM.

SALAD-Pan: Sensor-Agnostic Latent Adaptive Diffusion for Pan-Sharpening

Recently, diffusion models bring novel insights for Pan-sharpening and notably boost fusion precision. However, most existing models perform diffusion in the pixel space and train distinct models for different multispectral (MS) imagery, suffering from high latency and sensor-specific limitations. In this paper, we present SALAD-Pan, a sensor-agnostic latent space diffusion method for efficient pansharpening. Specifically, SALAD-Pan trains a band-wise single-channel VAE to encode high-resolution multispectral (HRMS) into compact latent representations, supporting MS images with various channel counts and establishing a basis for acceleration. Then spectral physical properties, along with PAN and MS images, are injected into the diffusion backbone through unidirectional and bidirectional interactive control structures respectively, achieving high-precision fusion in the diffusion process. Finally, a lightweight cross-spectral attention module is added to the central layer of diffusion model, reinforcing spectral connections to boost spectral consistency and further elevate fusion precision. Experimental results on GaoFen-2, QuickBird, and WorldView-3 demonstrate that SALAD-Pan outperforms state-of-the-art diffusion-based methods across all three datasets, attains a 2-3x inference speedup, and exhibits robust zero-shot (cross-sensor) capability.