OmniCapBench is a new benchmark for evaluating fine-grained audio-visual captioning, redefining evaluation as a structured diagnostic framework. It uses atomic, verifiable units across three tracks to assess entity references, visual shots, and audio events, enabling precise scoring and identifying perception errors in multimodal large language models (MLLMs).
The chart will appear after repeat observations. The current metric comes from the source.
2 source votesReal observations only. History before source connection is not reconstructed.
This framework allows for precise identification of multimodal perception errors, providing a detailed roadmap for improving audio-visual reasoning in large language models.
The paper introduces Trace2Env, a learning-free framework that uses historical interaction traces to create a reusable environment worldbook for simulating realistic environments without rebuilding the original system. It improves next-observation fidelity and long-horizon interaction consistency compared to conventional prompt-based language world models.
4 days agoDaily PapersCheckerBench is a benchmark for evaluating coding agents' ability to synthesize static-analysis checkers from defect specifications, including tasks derived from 300 CVEs across multiple repositories and language ecosystems. CheckerLab, an evaluation framework, measures diagnostic contrast, patch localization, and tool use, with results showing that current agents struggle to develop reliable checkers, achieving a maximum Pass@1 of 45.33%.
3 days agoDaily PapersTetris3D is a generative framework for 3D scene reconstruction from a single image, ensuring objects are geometrically and physically coherent. It uses explicit conditioning on surrounding objects' geometry and physical relationships, along with a physics-based dataset called ComOb for training and evaluation.
2 days agoDaily PapersThe paper investigates on-policy distillation (OPD) in cross-tokenizer settings, focusing on alignment coverage and supervision reliability. It finds that strict 1:1 token alignment covers most student-generated tokens despite vocabulary mismatches, and that restricting reverse KL to a top-16 subset of shared vocabulary achieves comparable accuracy to full shared-vocabulary OPD, while adding span supervision reduces accuracy.
3 days agoDaily PapersNAMVIS is a diffusion-free framework for multi-view image synthesis that uses geometry-conditioned next-scale autoregression, predicting discrete visual tokens through coarse-to-fine scale steps and parallel sampling across target views. It outperforms diffusion-based methods in PSNR, SSIM, and LPIPS while being over three times faster.
6 days agoDaily PapersSparse attention is used to reduce latency in long-sequence generation, but existing methods can degrade generation quality at high sparsity. MC-Sparse is a training-free framework that selects individual key-value tokens and organizes similar queries into groups for efficient GPU execution.
4 days ago