This paper introduces Image Bundle Composition (IBC), shifting image retrieval from scoring individual photos to dynamically assembling cohesive image collections…
WebWorld uses a web browser as a world model to improve vision-language models’ ability to generate and fix web code,…
ContextBias introduces ContextBench, a benchmark of 92 professional roles and 1,656 controlled prompts, to test whether stereotypical associations between professions…
CoVA-SFT is a structured corpus of 51,900 samples containing over 222,000 multimodal reasoning steps, designed to train multimodal language models…
EvoGenUI-Bench evaluates LLMs’ ability to maintain and evolve interactive web interfaces across multiple turns, comprising 150 five-turn tasks across three…
SafeAtlas-VL is a dataset of 1.5 million annotated instances for multimodal safety evaluation that replaces binary classification with a five-level…
BLARM is a feed-forward method for generating animated 3D meshes from monocular video without explicit rigging or skeletal annotations, representing…
This paper studies whether function vectors — task-specific latent directions derived from in-context demonstrations — transfer across languages for emotion…
Chat-Edit-3D++ is a dialogue-based approach for editing 3D and 4D scenes that uses an LLM to interpret flexible text input…
RECAP-Forcing is a training-free inference method for long video generation that organizes memory by content novelty rather than temporal recency,…
This paper introduces Data Intrinsic Consistency (DIC), a self-scoring metric measuring sample-level inter-component consistency through Visual Information Consistency and Response…
SpanCalib-VLM combines a multimodal sequence tagger with a fine-tuned generative vision-language model to detect hallucinations in large VLMs, using a…
