On the Effect of Image Resolution on Semantic Segmentation
This addresses the problem of preserving fine details in semantic segmentation for applications like autonomous driving, though it appears incremental as it builds on existing resolution-handling techniques.
The paper tackles the computational challenge of high-resolution semantic segmentation by proposing a streamlined model that processes images at native resolution, matching the performance of more complex systems that use downscaling/upscaling. On the Cityscapes dataset, they achieve improved accuracy using Noisy Student Training.
High-resolution semantic segmentation requires substantial computational resources. Traditional approaches in the field typically downscale the input images before processing and then upscale the low-resolution outputs back to their original dimensions. While this strategy effectively identifies broad regions, it often misses finer details. In this study, we demonstrate that a streamlined model capable of directly producing high-resolution segmentations can match the performance of more complex systems that generate lower-resolution results. By simplifying the network architecture, we enable the processing of images at their native resolution. Our approach leverages a bottom-up information propagation technique across various scales, which we have empirically shown to enhance segmentation accuracy. We have rigorously tested our method using leading-edge semantic segmentation datasets. Specifically, for the Cityscapes dataset, we further boost accuracy by applying the Noisy Student Training technique.