Benchmarking Feature Upsampling Methods for Vision Foundation Models using Interactive Segmentation
This work addresses the challenge of adapting VFMs for dense prediction tasks like interactive segmentation, which is incremental as it focuses on benchmarking existing upsampling methods rather than introducing new ones.
The paper tackled the problem of low-resolution features in Vision Foundation Models (VFMs) limiting dense prediction tasks by benchmarking feature upsampling methods using Interactive Segmentation as a novel evaluation framework, showing that appropriate upsampling strategies significantly improve VFM feature quality.
Vision Foundation Models (VFMs) are large-scale, pre-trained models that serve as general-purpose backbones for various computer vision tasks. As VFMs' popularity grows, there is an increasing interest in understanding their effectiveness for dense prediction tasks. However, VFMs typically produce low-resolution features, limiting their direct applicability in this context. One way to tackle this limitation is by employing a task-agnostic feature upsampling module that refines VFM features resolution. To assess the effectiveness of this approach, we investigate Interactive Segmentation (IS) as a novel benchmark for evaluating feature upsampling methods on VFMs. Due to its inherent multimodal input, consisting of an image and a set of user-defined clicks, as well as its dense mask output, IS creates a challenging environment that demands comprehensive visual scene understanding. Our benchmarking experiments show that selecting appropriate upsampling strategies significantly improves VFM features quality. The code is released at https://github.com/havrylovv/iSegProbe