Object Instance Identification in Dynamic Environments

TL;DR

This paper addresses the challenge of identifying object instances in dynamic environments with appearance changes due to interactions, occlusions, and background variations, by creating a new benchmark and analyzing key factors affecting performance.

Contribution

The authors built a new benchmark with over 1,500 instances on the EPIC-KITCHENS dataset and analyzed the challenges of object instance identification in dynamic, interactive environments.

Findings

Robustness against appearance changes is crucial.

Combining low-level and high-level features improves identification.

Foreground feature selection helps distinguish overlapping objects.

Abstract

We study the problem of identifying object instances in a dynamic environment where people interact with the objects. In such an environment, objects' appearance changes dynamically by interaction with other entities, occlusion by hands, background change, etc. This leads to a larger intra-instance variation of appearance than in static environments. To discover the challenges in this setting, we newly built a benchmark of more than 1,500 instances built on the EPIC-KITCHENS dataset which includes natural activities and conducted an extensive analysis of it. Experimental results suggest that (i) robustness against instance-specific appearance change (ii) integration of low-level (e.g., color, texture) and high-level (e.g., object category) features (iii) foreground feature selection on overlapping objects are required for further improvement.