Monocular Depth Estimation on Jaehun's Bloghttps://jaehun.me/en/tags/monocular-depth-estimation/Recent content in Monocular Depth Estimation on Jaehun's BlogHugoen-USSun, 13 Sep 2026 09:29:41 +0900Marigold V2: Revisiting Diffusion Transformers for Monocular Depth Estimationhttps://jaehun.me/en/posts/marigold-v2-revisiting-diffusion-transformers-for-monocular-depth-estimation/Sun, 13 Sep 2026 00:00:00 +0900https://jaehun.me/en/posts/marigold-v2-revisiting-diffusion-transformers-for-monocular-depth-estimation/<p><a href="https://arxiv.org/abs/2609.08084"target="_blank" class="inline-flex items-center gap-1" >Paper<svg class="h-3 w-3 flex-shrink-0" id="external-link" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><path fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M15 3h6v6m-11 5L21 3m-3 10v6a2 2 0 0 1-2 2H5a2 2 0 0 1-2-2V8a2 2 0 0 1 2-2h6"/></svg> </a></p> <h2 id="marigold-v2-reviving-an-image-editing-diffusion-transformer-dit-as-a-monocular-depth-estimator">Marigold V2: Reviving an Image-Editing Diffusion Transformer (DiT) as a Monocular Depth Estimator<a href="#marigold-v2-reviving-an-image-editing-diffusion-transformer-dit-as-a-monocular-depth-estimator" class="heading-anchor" aria-label="Link to this section"><svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/><path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/></g></svg></a></h2><h2 id="tldr">TL;DR<a href="#tldr" class="heading-anchor" aria-label="Link to this section"><svg class="h-4 w-4" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" viewBox="0 0 24 24"><g fill="none" stroke="currentColor" stroke-linecap="round" stroke-linejoin="round" stroke-width="2"><path d="M10 13a5 5 0 0 0 7.54.54l3-3a5 5 0 0 0-7.07-7.07l-1.72 1.71"/><path d="M14 11a5 5 0 0 0-7.54-.54l-3 3a5 5 0 0 0 7.07 7.07l1.71-1.71"/></g></svg></a></h2><p>This study fine-tunes an <strong>image-editing diffusion transformer (DiT)</strong>, Qwen-Image-Edit-2509, with 4-bit QLoRA to achieve SOTA monocular depth estimation <strong>in about a week on a single 32GB GPU</strong>. The key is two new losses — <strong>iREPA-depth</strong>, which aligns with semantic features extracted from depth GT, and <strong>SinkLoss</strong> based on optimal transport (Sinkhorn) — which suppress flying-pixel artifacts while preserving fine details such as fur, foliage, and hair.</p>