AI Paper Atlas

Research map / 3D vision / Depth and stereo

FiffDepth: Feed-forward Transformation of Diffusion-Based Generators for Detailed Depth Estimation

Yunpeng Bai, Qixing Huang

ICCV 2025 · 3D vision › Depth and stereo · approach: Self-supervised monocular depth · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
keep the diffusion trajectory with blended image/depth latents during training (ablation)FiffDepth · Depth benchmarks (qualitative)detail preservationImproved
predict blended image-depth latent rather than purely image latents along the trajectory (ablation)FiffDepth · Depth benchmarks (qualitative)relative depth relationshipsImproved
add DINOv2 pseudo-label supervision via learnable filter on real images (ablation)FiffDepth · Depth benchmarks (qualitative)generalizationImproved
apply DINO supervision at d_0 instead of filtered d_-1 (ablation)FiffDepth · Depth benchmarks (qualitative)detail preservationHurt
use feed-forward single-step inference instead of multi-step stochastic diffusionFiffDepth vs Marigold · Inference timetime (s): 103 -> 0.092Improved
use FiffDepth instead of Depth Pro for boundariesFiffDepth · Sintel boundary F1F1: 0.409 -> 0.423Improved
use FiffDepth instead of Marigold on ETH3DFiffDepth · ETH3DAbsRel: 6.4 -> 7.1Hurt
use FiffDepth instead of DepthAnything v2-L on ETH3DFiffDepth · ETH3DAbsRel: 13.1 -> 7.1Improved
use FiffDepth instead of DepthAnything v2-L on DA-2KFiffDepth · DA-2Kaccuracy: 97.1 -> 97.1No clear effect
use FiffDepth instead of Marigold on NYUv2FiffDepth · NYUv2AbsRel: 5.5 -> 4.4Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy