跳到正文
研踪PaperTrail
← 研踪 PaperTrail

arXiv:2611.17903 · v1

Zoom In, Scale Out: Metric Depth Models Mistake Digital Zoom for Moving Closer

首次提交 2026-11-13 16:47
版本更新 2026-11-13 16:47
雷达观测 2026-11-16 10:40
arXiv ↗PDF ↗

摘要

Many new AI models estimate how far away things are, in metres, from a single photo. To do this they must also work out what camera took the photo. We check whether they really do with a simple test: digital zoom, that is, cropping a photo and enlarging the crop back to full size. Zooming moves nothing in the scene, so a correct model should give the same distance for every point still in the picture. The test is free, exact and needs no labels. Two numbers describe a model's reaction: its depth changes like s−αs^{-α} (ideal α=0α=0) and its focal-length estimate like sβs^β (ideal β=1β=1), for zoom factor ss. On 800 images from KITTI, NYUv2 and DIODE, the widely used Depth Anything V2 has α=0.95α=0.95: it treats a zoomed photo almost as if the camera had moved closer, because it judges distance mainly from how large things look. Models that also predict the camera remove only part of this error (α≈0.31α\approx0.31-0.360.36), because their focal estimate grows too little. Given the true camera, Depth Pro and DA3-Nested keep their depth (α≈0α\approx0), so the error lies in their camera estimate. The reason: a zoomed photo is, pixel for pixel, a normal photo of a scene squashed in depth, so size alone cannot tell the two apart, and a focal estimator trained by regression stays close to the cameras it saw in training. Undoing this needs no labels or training and makes Depth Pro pass the test (α=0.04α=0.04) at almost no cost on normal photos. Controls, including a real optical zoom lens, rule out blur and resizing artefacts. Finally, label-free fine-tuning for self-consistency lowers αα to 0.05 on KITTI and 0.07 on NYUv2 once a "blurry means zoomed" shortcut is removed; we report what this costs on normal and unseen photos.

作者备注:22 pages including supplementary material. Code: https://github.com/rabeyanoor/zoom-equivariance

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

目前只收录了一个版本。收录下一版后,可以对照标题、作者、分类和摘要。

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。

v1 · 2026-11-13 16:47 · 最新观测版本

Zoom In, Scale Out: Metric Depth Models Mistake Digital Zoom for Moving Closer

Many new AI models estimate how far away things are, in metres, from a single photo. To do this they must also work out what camera took the photo. We check whether they really do with a simple test: digital zoom, that is, cropping a photo and enlarging the crop back to full size. Zooming moves nothing in the scene, so a correct model should give the same distance for every point still in the picture. The test is free, exact and needs no labels. Two numbers describe a model's reaction: its depth changes like s−αs^{-α} (ideal α=0α=0) and its focal-length estimate like sβs^β (ideal β=1β=1), for zoom factor ss. On 800 images from KITTI, NYUv2 and DIODE, the widely used Depth Anything V2 has α=0.95α=0.95: it treats a zoomed photo almost as if the camera had moved closer, because it judges distance mainly from how large things look. Models that also predict the camera remove only part of this error (α≈0.31α\approx0.31-0.360.36), because their focal estimate grows too little. Given the true camera, Depth Pro and DA3-Nested keep their depth (α≈0α\approx0), so the error lies in their camera estimate. The reason: a zoomed photo is, pixel for pixel, a normal photo of a scene squashed in depth, so size alone cannot tell the two apart, and a focal estimator trained by regression stays close to the cameras it saw in training. Undoing this needs no labels or training and makes Depth Pro pass the test (α=0.04α=0.04) at almost no cost on normal photos. Controls, including a real optical zoom lens, rule out blur and resizing artefacts. Finally, label-free fine-tuning for self-consistency lowers αα to 0.05 on KITTI and 0.07 on NYUv2 once a "blurry means zoomed" shortcut is removed; we report what this costs on normal and unseen photos.

查看此版本解读 →查看此版本原文 ↗