Research map / Label-efficient and robust learning / Domain adaptation and generalization
Museum Exhibit Identification Challenge for the Supervised Domain Adaptation and Beyond
Piotr Koniusz, Yusuf Tas, Hongguang Zhang, Mehrtash Harandi, Fatih Porikli, Rui Zhang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Euclidean alignment with JBLD distance (ablation) | So-HoT supervised domain adaptation · Open MIC combined set | top-1 accuracy (%): 62.8 -> 64.2 | Improved |
| add target-stream fine-tuning with JBLD alignment versus S+T | VGG16 two-stream · Open MIC combined set | top-1 accuracy (%): 55.9 -> 64.2 | Improved |
| replace Euclidean alignment with JBLD distance (ablation) | So-HoT · Office AD, VGG16 | accuracy (%): 89.45 -> 90.80 | Improved |
| replace Euclidean alignment with AIRM distance (ablation) | So-HoT · Office AD, VGG16 | accuracy (%): 89.45 -> 90.72 | Improved |
| replace Euclidean alignment with JBLD distance (ablation) | So-HoT · Open MIC Clk subset | accuracy (%): 59.3 -> 61.2 | Improved |
| use AIRM instead of JBLD distance (ablation) | So-HoT · Open MIC Clk subset | accuracy (%): 61.2 vs 61.1 | No clear effect |
| switch backbone from VGG16 to GoogLeNet (ablation) | two-stream JBLD · Office AD | accuracy (%): 90.80 -> 91.33 | Improved |
| add source data to target-only fine-tuning (S+T vs T) | VGG16 · Open MIC combined set | top-1 accuracy (%): 56.5 -> 55.9 | Hurt |
| project features with Nystrom projections before non-Euclidean distance (ablation) | JBLD/AIRM alignment · training time | speed: 210x faster than naive non-Euclidean distances | Improved |
| add JBLD domain adaptation on top of S+T | VGG16 · Open MIC subset Shx | top-1 accuracy (%): 29.0 -> 48.5 | Improved |
| add JBLD domain adaptation over source-only (S) | VGG16 · Open MIC subset Gls | top-1 accuracy (%): 15.8 -> 74.4 | Improved |
| replace Euclidean alignment with JBLD distance (ablation) | So-HoT · Office-Home Cl->Ar | accuracy (%): 48.1 -> 49.2 | Improved |
Closest papers by meaning
- The Met Dataset: Instance-level Recognition for Artworks (NeurIPS 2021 · similarity 0.89)
- Taming CLIP for Fine-grained and Structured Visual Understanding of Museum Exhibits (ECCV 2024 · similarity 0.85)
- Understanding Museum Exhibits using Vision-Language Reasoning (ICCV 2025 · similarity 0.84)
- Tools Identification By On-Board Adaptation of Vision-and-Language Models (AAAI 2024 · similarity 0.81)
- PoSh: Using Scene Graphs to Guide LLMs-as-a-Judge for Detailed Image Descriptions (ICLR 2026 · similarity 0.81)
- Semi-Truths: A Large-Scale Dataset of AI-Augmented Images for Evaluating Robustness of AI-Generated Image detectors (NeurIPS 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
