BIMBA: Selective-Scan Compression for Long-Range Video Question Answering
This work addresses the problem of computational inefficiency and information loss in long-form VQA for researchers and practitioners, offering a novel compression method that improves accuracy over prior approaches.
The paper tackles the challenge of extracting relevant information and modeling long-range dependencies in long videos for Video Question Answering (VQA) by introducing BIMBA, an efficient state-space model that uses selective scanning to compress video data, achieving state-of-the-art accuracy on multiple benchmarks such as PerceptionTest and NExT-QA.
Video Question Answering (VQA) in long videos poses the key challenge of extracting relevant information and modeling long-range dependencies from many redundant frames. The self-attention mechanism provides a general solution for sequence modeling, but it has a prohibitive cost when applied to a massive number of spatiotemporal tokens in long videos. Most prior methods rely on compression strategies to lower the computational cost, such as reducing the input length via sparse frame sampling or compressing the output sequence passed to the large language model (LLM) via space-time pooling. However, these naive approaches over-represent redundant information and often miss salient events or fast-occurring space-time patterns. In this work, we introduce BIMBA, an efficient state-space model to handle long-form videos. Our model leverages the selective scan algorithm to learn to effectively select critical information from high-dimensional video and transform it into a reduced token sequence for efficient LLM processing. Extensive experiments demonstrate that BIMBA achieves state-of-the-art accuracy on multiple long-form VQA benchmarks, including PerceptionTest, NExT-QA, EgoSchema, VNBench, LongVideoBench, and Video-MME. Code, and models are publicly available at https://sites.google.com/view/bimba-mllm.