---
format: "aidr-story-markdown/v1"
id: "329b53997a0b53943e4c2870c19a92f9bb89b30c340d8ee5962293687422c5b8"
canonical_url: "https://aidr.today/329b5399?lang=en"
title: "Policy-Mask: Budget-Aware Adaptation of Diffusion Features to Discriminative Tasks"
lang: "en"
requested_lang: "en"
available_langs: ["en","vi"]
translation_fallback: null
fallback_fields: []
published_at: "2026-10-23T00:00:00.000Z"
category: "Research"
topics: ["inference"]
source_urls: ["https://huggingface.co/papers/2610.28889"]
summary: "Policy-Mask is a budget-conditioned policy that jointly selects timestep, block, and channel subset from a frozen diffusion model for discriminative tasks. It improves over fixed extraction and feature-filter baselines across segmentation, classification, and semantic correspondence, reaching 38.9 mIoU on ADE20K, 75.4% top-1 on CIFAR-100 at 128 dimensions, and 76.3 PCK@0.1 on SPair-71k."
---

# Policy\-Mask: Budget\-Aware Adaptation of Diffusion Features to Discriminative Tasks

> [Open the canonical story](<https://aidr.today/329b5399?lang=en>)

**Published:** 2026-10-23T00:00:00.000Z
**Category:** Research
**Topics:** inference

## Summary

Policy\-Mask is a budget\-conditioned policy that jointly selects timestep, block, and channel subset from a frozen diffusion model for discriminative tasks\. It improves over fixed extraction and feature\-filter baselines across segmentation, classification, and semantic correspondence, reaching 38\.9 mIoU on ADE20K, 75\.4% top\-1 on CIFAR\-100 at 128 dimensions, and 76\.3 PCK@0\.1 on SPair\-71k\.

## Sources

- [Story source](<https://huggingface.co/papers/2610.28889>)