Unsupervised Object Detection with Theoretical Guarantees
This work addresses the lack of theoretical guarantees in unsupervised object detection for researchers and practitioners, representing a novel foundational advancement rather than an incremental improvement.
The authors tackled the problem of unsupervised object detection by developing a method with theoretical guarantees to recover true object positions up to quantifiable small shifts, validated through synthetic experiments with pixel-level precision and CLEVR-based data showing errors within theoretical bounds.
Unsupervised object detection using deep neural networks is typically a difficult problem with few to no guarantees about the learned representation. In this work we present the first unsupervised object detection method that is theoretically guaranteed to recover the true object positions up to quantifiable small shifts. We develop an unsupervised object detection architecture and prove that the learned variables correspond to the true object positions up to small shifts related to the encoder and decoder receptive field sizes, the object sizes, and the widths of the Gaussians used in the rendering process. We perform detailed analysis of how the error depends on each of these variables and perform synthetic experiments validating our theoretical predictions up to a precision of individual pixels. We also perform experiments on CLEVR-based data and show that, unlike current SOTA object detection methods (SAM, CutLER), our method's prediction errors always lie within our theoretical bounds. We hope that this work helps open up an avenue of research into object detection methods with theoretical guarantees.