On Discriminative vs. Generative classifiers: Rethinking MLLMs for Action Understanding

TL;DR

This paper compares generative and discriminative classifiers in multimodal large language models for action understanding, proposing a hybrid GAD approach that improves accuracy and efficiency over existing methods.

Contribution

It introduces the GAD classifier that combines generative and discriminative approaches, enhancing performance while maintaining efficiency in action understanding tasks.

Findings

Discriminative classifiers outperform generative ones in accuracy and efficiency.

GAD improves accuracy by 2.5% and triples inference speed on COIN benchmark.

GAD achieves state-of-the-art results across multiple datasets.

Abstract

Multimodal Large Language Models (MLLMs) have advanced open-world action understanding and can be adapted as generative classifiers for closed-set settings by autoregressively generating action labels as text. However, this approach is inefficient, and shared subwords across action labels introduce semantic overlap, leading to ambiguity in generation. In contrast, discriminative classifiers learn task-specific representations with clear decision boundaries, enabling efficient one-step classification without autoregressive decoding. We first compare generative and discriminative classifiers with MLLMs for closed-set action understanding, revealing the superior accuracy and efficiency of the latter. To bridge the performance gap, we design strategies that elevate generative classifiers toward performance comparable with discriminative ones. Furthermore, we show that generative modeling can complement discriminative classifiers, leading to better performance while preserving efficiency. To this end, we propose Generation-Assisted Discriminative~(GAD) classifier for closed-set action understanding. GAD operates only during fine-tuning, preserving full compatibility with MLLM pretraining. Extensive experiments on temporal action understanding benchmarks demonstrate that GAD improves both accuracy and efficiency over generative methods, achieving state-of-the-art results on four tasks across five datasets, including an average 2.5% accuracy gain and 3x faster inference on our largest COIN benchmark.