Scene Change Detection

12 benchmarks25 papers

Scene change detection (SCD) refers to the task of localizing changes and identifying change-categories given two scenes. A scene can be either an RGB (+D) image or a 3D reconstruction (point cloud). If the scene is an image, SCD is a form of pixel-level prediction because each pixel in the image is classified according to a category. On the other hand, if the scene is point cloud, SCD is a form of point-level prediction because each point in the cloud is classified according to a category.

Some example benchmarks for this task are VL-CMU-CD, PCD, and CD2014. Recently, more complicated benchmarks such as ChangeSim, HDMap, and Mallscape are released.

Models are usually evaluated with the Mean Intersection-Over-Union (Mean IoU), Pixel Accuracy, or F1 metrics.

Benchmarks

Scene Change Detection on ChangeSim

Category mIoUmacro F1

Scene Change Detection on PCD

F1-score

Scene Change Detection on Unaligned-VL-CMU-CD (neighbor distance 2)

F1-score

Scene Change Detection on VL-CMU-CD

F1-score

Scene Change Detection on ChangeVPR

F1 score

Scene Change Detection on Semantic Scene Understanding Challenge (passive actuation & ground-truth localisation)

OMQavg_pairwiseavg_labelavg_spatialavg_state_qualityavg_fp_quality