Face Density as a Proxy for Data Complexity: Quantifying the Hardness of Instance Count
For practitioners in computer vision, this work provides a rigorous, controlled demonstration that instance density is an intrinsic dimension of data hardness, motivating density-aware training and evaluation strategies.
The paper quantifies how increasing face density (number of faces per image) monotonically degrades model performance across classification, regression, and detection tasks, with error rates increasing up to 4.6x when models trained on low-density images are tested on high-density ones.
Machine learning progress has historically prioritized model-centric innovations, yet achievable performance is frequently capped by the intrinsic complexity of the data itself. In this work, we isolate and quantify the impact of instance density (measured by face count) as a primary driver of data complexity. Rather than simply observing that ``crowded scenes are harder,'' we rigorously control for class imbalance to measure the precise degradation caused by density alone. Controlled experiments on the WIDER FACE and Open Images datasets, restricted to exactly 1 to 18 faces per image with perfectly balanced sampling, reveal that model performance degrades monotonically with increasing face count. This trend holds across classification, regression, and detection paradigms, even when models are fully exposed to the entire density range. Furthermore, we demonstrate that models trained on low-density regimes fail to generalize to higher densities, exhibiting a systematic under-counting bias, with error rates increasing by up to 4.6x, which suggests density acts as a domain shift. These findings establish instance density as an intrinsic, quantifiable dimension of data hardness and motivate specific interventions in curriculum learning and density-stratified evaluation.