Research map / Label-efficient and robust learning / Domain adaptation and generalization
How Stable Are Transferability Metrics Evaluations?
Andrea Agostinelli, Michal Pándy, Jasper Uijlings, Thomas Mensink, Vittorio Ferrari
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 |
|---|---|---|---|
| vary the target dataset in the experimental setup | transferability metric evaluation · Scenario 1 segmentation | Setup Stability: 0.23 vs 0.53 (E) and 0.60 (S) | Hurt |
| vary the evaluation measure instead of the target | transferability metric evaluation · Scenario 1 segmentation | Setup Stability: 0.53 | Mixed |
| vary the source pool | transferability metric evaluation · Scenario 1 segmentation | Setup Stability: 0.60 | Improved |
| vary the target dataset | transferability metric evaluation · Scenario 2 classification | Setup Stability: 0.62 vs 0.69 (E) and 0.80 (S) | Hurt |
| aggregate outcomes across many experiments | LogME · Scenario 1 segmentation | win rate (%): 56 | Improved |
| aggregate outcomes across many experiments | NLEEP · Scenario 2 classification | win rate (%): 73 | Improved |
| switch from segmentation source datasets to classification source architectures | NLEEP · Scenario 1 vs 2 | win rate (%): 4 -> 73 | Improved |
| switch from segmentation source datasets to classification source architectures | LogME · Scenario 1 vs 2 | win rate (%): 56 -> 17 | Hurt |
| sample a variable 2-100% number of target classes instead of fixed 50% | LogME · Bonus scenario average | weighted Kendall tau: 0.462 -> -0.781 | Hurt |
| sample a variable number of target classes (2-100%) instead of fixed 50% | GBC · Bonus scenario average | weighted Kendall tau: 0.543 -> 0.898 | Improved |
| sample variable class counts; trivial NumC baseline | NumC · Bonus scenario (2-100% classes) | weighted Kendall tau: 0.866 vs GBC 0.898 | Improved |
| sample a variable number of target classes | H-score · Bonus scenario average | weighted Kendall tau: 0.250 -> 0.179 | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LEEP — LEEP: A New Measure to Evaluate Transferability of Learned Representations (ICML 2020) · better on 4 of 4 shared results
- LogME — LogME: Practical Assessment of Pre-trained Models for Transfer Learning (ICML 2021) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Transferability Metrics for Selecting Source Model Ensembles (CVPR 2022 · similarity 0.90)
- How NOT to benchmark your SITE metric: Beyond Static Leaderboards and Towards Realistic Evaluation. (ICLR 2026 · similarity 0.88)
- Building a Winning Team: Selecting Source Model Ensembles using a Submodular Transferability Estimation Approach (ICCV 2023 · similarity 0.88)
- Transferability Estimation Using Bhattacharyya Class Separability (CVPR 2022 · similarity 0.87)
- Implicit Modeling for Transferability Estimation of Vision Foundation Models (NeurIPS 2025 · similarity 0.83)
- Understanding the Transferability of Representations via Task-Relatedness (NeurIPS 2024 · similarity 0.82)
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.
