Monitored Distillation for Positive Congruent Depth Completion

TL;DR

This paper introduces Monitored Distillation, a novel adaptive knowledge distillation method for depth completion that selectively learns from multiple teacher models without ground truth, improving accuracy and reducing model size.

Contribution

It proposes a positive congruent training approach called Monitored Distillation that leverages a confidence monitor to selectively distill depth predictions from multiple teachers without ground truth supervision.

Findings

Outperforms blind ensembling baselines by 17.53% on VOID

Outperforms unsupervised methods by 24.25% on VOID

Achieves 79% model size reduction while maintaining performance

Abstract

We propose a method to infer a dense depth map from a single image, its calibration, and the associated sparse point cloud. In order to leverage existing models (teachers) that produce putative depth maps, we propose an adaptive knowledge distillation approach that yields a positive congruent training process, wherein a student model avoids learning the error modes of the teachers. In the absence of ground truth for model selection and training, our method, termed Monitored Distillation, allows a student to exploit a blind ensemble of teachers by selectively learning from predictions that best minimize the reconstruction error for a given image. Monitored Distillation yields a distilled depth map and a confidence map, or ``monitor'', for how well a prediction from a particular teacher fits the observed image. The monitor adaptively weights the distilled depth where if all of the teachers exhibit high residuals, the standard unsupervised image reconstruction loss takes over as the supervisory signal. On indoor scenes (VOID), we outperform blind ensembling baselines by 17.53% and unsupervised methods by 24.25%; we boast a 79% model size reduction while maintaining comparable performance to the best supervised method. For outdoors (KITTI), we tie for 5th overall on the benchmark despite not using ground truth. Code available at: https://github.com/alexklwong/mondi-python.