AV-Reasoner: Improving and Benchmarking Clue-Grounded Audio-Visual Counting for MLLMs
This addresses a specific limitation in video understanding for MLLMs, though it is incremental as it builds on existing methods with new training techniques and a benchmark.
The paper tackles the problem of multimodal large language models (MLLMs) struggling with counting tasks by introducing CG-AV-Counting, a manually-annotated benchmark with 1,027 questions and 5,845 clues over 497 long videos, and proposes AV-Reasoner, which achieves state-of-the-art results on multiple benchmarks.
Despite progress in video understanding, current MLLMs struggle with counting tasks. Existing benchmarks are limited by short videos, close-set queries, lack of clue annotations, and weak multimodal coverage. In this paper, we introduce CG-AV-Counting, a manually-annotated clue-grounded counting benchmark with 1,027 multimodal questions and 5,845 annotated clues over 497 long videos. It supports both black-box and white-box evaluation, serving as a comprehensive testbed for both end-to-end and reasoning-based counting. To explore ways to improve model's counting capability, we propose AV-Reasoner, a model trained with GRPO and curriculum learning to generalize counting ability from related tasks. AV-Reasoner achieves state-of-the-art results across multiple benchmarks, demonstrating the effectiveness of reinforcement learning. However, experiments show that on out-of-domain benchmarks, reasoning in the language space fails to bring performance gains. The code and benchmark have been released on https://av-reasoner.github.io.