サインイン
インサイト - Fine-Grained Open-Vocabulary Object Detection
Limitations of CLIP in Fine-Grained Open-World Perception
The limitations observed in open-vocabulary object detectors regarding fine-grained understanding can be attributed to deficiencies within the CLIP latent space, where fine-grained object properties are not effectively captured by the standard cosine similarity matching.
Evaluating Open-Vocabulary Object Detectors' Ability to Discern Fine-Grained Object Properties
Current open-vocabulary object detectors struggle to accurately capture and distinguish fine-grained object details like color, material, pattern, and transparency.
Enhancing Fine-Grained Attribute Detection in Open-Vocabulary Object Recognition through Explicit Linear Composition
The core message of this paper is to propose a universal and explicit approach that enhances the fine-grained attribute detection capabilities of mainstream open-vocabulary object detection (OVD) models by highlighting fine-grained attributes in an explicit linear space.
