跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2609.06490 · v2

OracleZoom: Reference-Constrained Recursive Image Super-Resolution Beyond Ground Truth

首次提交 2026-09-06 17:06
版本更新 2026-11-18 13:32
雷达观测 2026-11-19 19:08
arXiv ↗PDF ↗

摘要

Recursive image Super-Resolution (SR) reaches extreme magnification by feeding each prediction back into the same model, like zooming into an image again and again. Real ground truth for deep zooms is impractical to capture (a full 256×256\times target would be a 52 GB image), so predictions at deeper zooms have no supervision. Yet the last ground-truth image still contains the region that every deeper zoom magnifies. OracleZoom turns this region into a training signal. Inspired by on-policy self-distillation, the model trains on its own prediction for one recursive step (4×→16×4\times \rightarrow 16\times), and a cross-scale loss projects each 16×16\times prediction back to the last ground truth at 4×4\times. This projection can verify only coarse content, so a no-reference quality objective guides the remaining fine detail, and a latent prior stops this objective from adding hallucinated artifacts. On seven datasets, OracleZoom achieves the highest CLIPIQA at every zoom depth (mean 0.713 vs. 0.621 for Chain-of-Zoom), with the largest gains at the deepest zooms, even though training stops at 16×16\times and optimizes a different quality metric. Pooled over two ground-truth-anchored judges, this added detail comes at the same hallucination rate as Chain-of-Zoom at 16×16\times and a lower rate at 4×4\times, and removing the cross-scale loss raises hallucination at 16×16\times.

作者备注:Project page: https://dipta007.github.io/OracleZoom/

AI 论文解读

基于摘要 · 非全文审阅

此版本尚未生成 AI 解读。解读由后台任务生成并按论文版本保存,登录后共享已有结果。

版本差异

本站首次收录为 v2,暂无更早版本可对照。

观测到的版本

相邻记录可能跳过中间版本;首次发现 v2 及以上也会标为版本更新。

v2 · 2026-11-18 13:32 · 最新观测版本

OracleZoom: Reference-Constrained Recursive Image Super-Resolution Beyond Ground Truth

Recursive image Super-Resolution (SR) reaches extreme magnification by feeding each prediction back into the same model, like zooming into an image again and again. Real ground truth for deep zooms is impractical to capture (a full 256×256\times target would be a 52 GB image), so predictions at deeper zooms have no supervision. Yet the last ground-truth image still contains the region that every deeper zoom magnifies. OracleZoom turns this region into a training signal. Inspired by on-policy self-distillation, the model trains on its own prediction for one recursive step (4×→16×4\times \rightarrow 16\times), and a cross-scale loss projects each 16×16\times prediction back to the last ground truth at 4×4\times. This projection can verify only coarse content, so a no-reference quality objective guides the remaining fine detail, and a latent prior stops this objective from adding hallucinated artifacts. On seven datasets, OracleZoom achieves the highest CLIPIQA at every zoom depth (mean 0.713 vs. 0.621 for Chain-of-Zoom), with the largest gains at the deepest zooms, even though training stops at 16×16\times and optimizes a different quality metric. Pooled over two ground-truth-anchored judges, this added detail comes at the same hallucination rate as Chain-of-Zoom at 16×16\times and a lower rate at 4×4\times, and removing the cross-scale loss raises hallucination at 16×16\times.

查看此版本解读 →查看此版本原文 ↗