Unverified50% confidenceOpinionExact time
直接用冻结的 DINOv2 全局特征跑 k-NN 是对其最不公平的评估方式，因为测试的是未经调整的嵌入空间几何结构
1
Sources
50%
Confidence
Long-term
Relevance
7/10/2026
First Seen
Sources
DINOv2 vs SigLIP: Why the Huge Gap in k-NN Classification? A Guide to Avoiding Vision Encoder Selection Pitfalls
redditr/MachineLearning7/8/2026
Related Claims
UnverifiedIn class-imbalanced scenarios, accuracy alone can be misleading and needs to be combined with a confusion matrix for more fine-grained evaluation.56% similarUnverifiedK-means依赖欧氏距离，在处理高基数类别特征时因维度灾难导致距离度量失效56% similarUnverifiedNex N2 Pro exhibits benchmark maxing (over-optimization on benchmarks), resulting in a significant gap between official benchmark results and actual performance.56% similarUnverifiedGridSearchCV将参数网格的穷举遍历与k折交叉验证结合，导致训练次数乘以k倍56% similarVerified若用含测试集的全量数据计算均值和标准差，会导致预处理泄漏，使评估指标虚高，无法真实反映模型泛化能力55% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/465000API
curl https://kongchang.com/api/v1/knowledge/claims/465000MCP
get_claim(id=465000)