Deep Value Networks Learn to Evaluate and Iteratively Refine Structured Outputs

TL;DR

This paper introduces Deep Value Networks (DVNs) that learn to evaluate structured outputs and iteratively refine them through gradient-based optimization, achieving state-of-the-art results in image segmentation and multi-label classification.

Contribution

The paper presents a novel deep value network framework that estimates task-specific loss functions and refines outputs via gradient descent, improving structured prediction performance.

Findings

Achieved state-of-the-art results on image segmentation benchmarks.

Outperformed existing methods in multi-label classification.

Demonstrated effective iterative refinement of structured outputs.

Abstract

We approach structured output prediction by optimizing a deep value network (DVN) to precisely estimate the task loss on different output configurations for a given input. Once the model is trained, we perform inference by gradient descent on the continuous relaxations of the output variables to find outputs with promising scores from the value network. When applied to image segmentation, the value network takes an image and a segmentation mask as inputs and predicts a scalar estimating the intersection over union between the input and ground truth masks. For multi-label classification, the DVN's objective is to correctly predict the F1 score for any potential label configuration. The DVN framework achieves the state-of-the-art results on multi-label prediction and image segmentation benchmarks.