PooDLe: Pooled and dense self-supervised learning from naturalistic videos
This work addresses the challenge of learning from minimally-curated, naturalistic video data for applications like driving and first-person video analysis, representing an incremental improvement over existing self-supervised methods.
The paper tackled the problem of self-supervised learning from naturalistic videos, which contain dense scenes and imbalanced data, by proposing PooDLe, a method combining pooled and dense objectives; results showed that a unified multi-scale objective is essential for learning effective representations, validated on BDD100K and Walking Tours datasets.
Self-supervised learning has driven significant progress in learning from single-subject, iconic images. However, there are still unanswered questions about the use of minimally-curated, naturalistic video data, which contain dense scenes with many independent objects, imbalanced class distributions, and varying object sizes. In this paper, we propose PooDLe, a self-supervised learning method that combines an invariance-based objective on pooled representations with a dense SSL objective that enforces equivariance to optical flow warping. Our results show that a unified objective applied at multiple feature scales is essential for learning effective image representations from naturalistic videos. We validate our method with experiments on the BDD100K driving video dataset and the Walking Tours first-person video dataset, demonstrating its ability to capture spatial understanding from a dense objective and semantic understanding via a pooled representation objective.