Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This paper introduces DensFiLM, a novel density-conditioned video saliency model that adapts to varying crowd densities by utilizing a Feature-wise Linear Modulation layer within a Video Swin Transformer architecture. By incorporating a learned density embedding, the model effectively adjusts channel-wise parameters to enhance saliency reconstruction based on the density of the scene, achieving significant performance improvements over existing models like ACLNet. The results demonstrate that this lightweight approach, which adds only ~100K parameters, outperforms traditional methods by leveraging density information, achieving mean NSS and CC scores of 1.434 and 0.517, respectively.
Density conditioning in video saliency models can yield performance gains without the need for larger model architectures, challenging the notion that bigger is always better.
Video saliency models typically apply a single fixation strategy across crowd scenes, despite systematic changes in attention with crowd density. Sparse scenes encourage tracking individuals, whereas dense scenes shift attention toward collective motion and scene-level landmarks. We introduce DensFiLM, a density-conditioned video saliency model that inserts a lightweight Feature-wise Linear Modulation layer at the bottleneck of a Video Swin Transformer. A learned density embedding produces channel-wise scale and shift parameters, allowing the decoder to reconstruct saliency from features selected for each density regime. The module adds only ~100K parameters and can use either CrowdFix density labels or the model's own density prediction. On CrowdFix, DensFiLM achieves mean NSS 1.434 and CC 0.517 over four seeds, improving over ACLNet by 14.7% and 14.9%, respectively, while predicted-density conditioning matches oracle-label performance. Ablations show that explicit RAFT optical flow and larger temporal and social-force extensions provide no further improvement in this setting. In a centre-prior-subtraction diagnostic, density conditioning yields an NSS gain of 0.462 over the unconditioned backbone, compared with 0.124 under standard evaluation. These results show that lightweight bottleneck conditioning provides a more effective inductive bias than increasing model capacity for crowd-video saliency. Our code is available at https://github.com/aniskhan25/crowdfix-saliency.