---
title: "Different Models, Shared Failures: Auditing Frozen-Encoder Pools for Medical Segmentation"
source: "arXiv AI/CL/LG（论文）"
source_tier: "t2"
url: "https://arxiv.org/abs/2610.41790v1"
permalink: "https://feed.astralor.com/items/itm_94d1d3676283d6a052cd8428"
published: "2026-10-30T14:21:52.000Z"
score: 57
channel: "tech"
tags: ["RAG","Different Models","Shared Failures","Auditing Frozen-Encoder Pools","Medical Segmentation Average","CoFailMed","Across","Average"]
---

# Different Models, Shared Failures: Auditing Frozen-Encoder Pools for Medical Segmentation

## At a glance

Average segmentation accuracy and model count do not reveal whether models fail on the same cases. We present CoFailMed, an audit of correlated failures in pretrained-encoder pools for medical segmentation. Across five …

## Key points

- Average segmentation accuracy and model count do not reveal whether models fail on the same cases.
- We present CoFailMed, an audit of correlated failures in pretrained-encoder pools for medical segmentation.
- Across five 2D tasks, we evaluate 11 public pretrained encoders (a controlled sample, not a census) using frozen features and separately trained $1\times 1$ readouts with four seeds per encoder.

## Full text

Average segmentation accuracy and model count do not reveal whether models fail on the same cases. We present CoFailMed, an audit of correlated failures in pretrained-encoder pools for medical segmentation. Across five 2D tasks, we evaluate 11 public pretrained encoders (a controlled sample, not a census) using frozen features and separately trained $1\times 1$ readouts with four seeds per encoder. Shared-checkpoint pairs have the highest mean per-case Dice correlation. An expanded 17-encoder analysis shows that distinct checkpoints within the same encoder family remain more correlated than cross-family pairs. At the pooled 33rd-percentile Dice threshold, four-member pools drawn from distinct encoders have mean effective sizes of 1.56-2.92, estimated from failure correlations in pools with enough failures and passes per member. In the primary pool, richer readouts and fine-tuning narrow the Dice-correlation gap between pairs sharing a checkpoint (or the DINOv2-B/S release) and all other pairs. Yet dependence between different encoders can increase, and fine-tuning does not consistently increase effective pool size at 33rd-percentile thresholds recomputed for each training recipe. A smaller between-group gap can coexist with greater dependence. Pool evaluation should therefore report absolute dependence and threshold-specific effective size alongside accuracy and nominal pool size. The CoFailMed release provides aligned per-case traces, analysis scripts and study code to reproduce the central audit results.

---

Verify against the original before citing; curated by Astral Feed: https://feed.astralor.com/items/itm_94d1d3676283d6a052cd8428