The paper introduces the Semantic Drift Protocol (SDP), a Telephone Game-inspired method to evaluate how well unified models preserve meaning when alternating between image-to-text (I2T) and text-to-image (T2I) generation over multiple generations. It defines Mean Cumulative Drift (MCD) and Multi-Generation GenEval (MGG) as metrics for semantic retention, and presents a new benchmark of 400 image‑text pairs from NoCaps and DOCCI to stress-test models beyond COCO. Applying SDP to seven models shows that models with strong single‑pass scores can still suffer severe semantic drift, revealing catastrophic failure modes that isolated benchmarks miss.
arXiv:2607. 22705v1 Announce Type: cross Abstract: Object-centric learning aims to represent scenes as objects whose properties can be reused in new combinations.
The paper introduces Dual-Stream Semantic Guidance (DSSG), a framework for Source‑Fully‑Free Domain Adaptation of Vision‑Language Models that mitigates dual semantic drift through a caption stream and a class‑anchor stream. It adds a Dynamic Cross‑Modal Knowledge Distillation module and a Prototype Anchor Calibration extension (DSSG‑PAC) to reduce computation while maintaining performance. Experiments show DSSG outperforms state‑of‑the‑art methods and DSSG‑PAC cuts adaptation time by 18.9% with minimal loss in accuracy.
TokenSwap introduces a method to benchmark and reduce the modality gap in multimodal large language models (MLLMs) by interleaving semantically aligned images with text tokens. By transforming text-only benchmarks like MMLU into image‑interleaved versions (TokenSwap‑Bench), the authors reveal a pervasive performance drop—averaging 19.6%—across 42 MLLMs, with reasoning models showing smaller gaps. Training with TokenSwap mitigates this gap while maintaining strong text‑only and vision‑language performance.
arXiv:2609.06245v1 Announce Type: cross Abstract: Multimodal Large Language Models (MLLMs) perform strongly on general visual understanding tasks such as visual question answering, yet they often str...
The paper introduces Auto-Comp, a fully automated, concept-driven pipeline that generates photorealistic compositional benchmarks for vision‑language models. Auto‑Comp creates paired Minimal and Contextual samples for each concept, enabling isolation of core binding abilities from visio‑linguistic complexity. Evaluations across 25 models reveal consistent failures in attribute and relational binding, with context helping relational tasks but hindering attribute tasks due to visual clutter.
arXiv:2603. 12055v3 Announce Type: replace-cross Abstract: Continual learning of pretrained vision-language models (VLMs) is prone to catastrophic forgetting, yet current approaches adapt to new tasks without explicitly preserving the cross-modal semantic geometry inherited from pretraining and previous stages, allowing new-task supervision to induce geometric distortion.
MinCU is a new benchmark for grounded minimal‑change understanding that presents pairs of near‑identical images differing by a single atomic variation in object category, attribute, count, or spatial position. Models are evaluated on their ability to describe the change, localize the changed region, and identify the changed entity. The authors also introduce SG‑ISA, a structured autoregressive method that decomposes the task into a Think‑Locate‑Describe sequence, showing that fine‑tuning with SG‑ISA improves both grounding accuracy and description quality while reducing reasoning‑token overhead.
arXiv:2607. 01813v1 Announce Type: cross Abstract: Evaluation benchmarks are essential for assessing vision-language models (VLMs), but most multimodal benchmarks are static, making them vulnerable to temporal staleness, data contamination, and costly maintenance.
arXiv:2608.22725v1 Announce Type: new Abstract: Short-drama generation has grown into a large, industrialized pipeline, and as it scales from isolated shots to the episode level, visual continuity ha...
STRAND is a new benchmark that tests multimodal large language models’ ability to track objects, their states, and relationships over time in videos. It evaluates intermediate reasoning by breaking queries into sub‑questions and uses Faithful Accuracy to ensure all parts of an answer are correct. The authors also propose an object‑centric framework that builds structured trajectories and shows reduced hallucinations and better temporal consistency compared to existing models.
arXiv:2606. 24716v1 Announce Type: cross Abstract: Sparse autoencoders (SAEs) are increasingly used to extract interpretable concepts from vision and vision language models, yet existing evaluation methods largely rely on proxy metrics or qualitative inspection rather than measuring semantic correspondence.
arXiv:2605. 18160v2 Announce Type: replace-cross Abstract: In recent years, multimodal large language models (MLLMs) have achieved remarkable progress, primarily attributed to effective paradigms for integrating visual and textual information.