InstAP: Instance-Aware Vision-Language Pre-Train for Spatial-Temporal Understanding

TL;DR

InstAP introduces an instance-aware vision-language pre-training framework that enhances spatial-temporal understanding by grounding textual mentions to specific regions, outperforming existing models in instance-level retrieval and zero-shot tasks.

Contribution

The paper proposes InstAP, a novel pre-training method with a large-scale dataset, enabling joint global and fine-grained instance-level alignment for improved vision-language understanding.

Findings

Outperforms existing VLP models on instance-level retrieval tasks.

Achieves competitive zero-shot performance on multiple video benchmarks.

Effectively localizes textual mentions to correct spatial-temporal instances.

Abstract

Current vision-language pre-training (VLP) paradigms excel at global scene understanding but struggle with instance-level reasoning due to global-only supervision. We introduce InstAP, an Instance-Aware Pre-training framework that jointly optimizes global vision-text alignment and fine-grained, instance-level contrastive alignment by grounding textual mentions to specific spatial-temporal regions. To support this, we present InstVL, a large-scale dataset (2 million images, 50,000 videos) with dual-granularity annotations: holistic scene captions and dense, grounded instance descriptions. On the InstVL benchmark, InstAP substantially outperforms existing VLP models on instance-level retrieval, and also surpasses a strong VLP baseline trained on the exact same data corpus, isolating the benefit of our instance-aware objective. Moreover, instance-centric pre-training improves global understanding: InstAP achieves competitive zero-shot performance on multiple video benchmarks, including MSR-VTT and DiDeMo. Qualitative visualizations further show that InstAP localizes textual mentions to the correct instances, while global-only models exhibit more diffuse, scene-level attention.