AI Paper Atlas

Research map / 3D vision / Depth and stereo

Monocular Relative Depth Perception With Web Stereo Data Supervision

Ke Xian, Chunhua Shen, Zhiguo Cao, Hao Lu, Yang Xiao, Ruibo Li, Zhenbo Luo

CVPR 2018 · 3D vision › Depth and stereo · approach: Self-supervised monocular depth · 225 citations (OpenAlex)

Read the paper

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
Train with ReDWeb pre-training followed by DIW fine-tuning rather than DIW training aloneMonocular relative depth model · DIW testWHDR: 14.98% -> 11.37%Improved
Train on ReDWeb with multiple point pairs rather than a single pair per imageMonocular relative depth model · DIW testWHDR: 22.14% -> 14.33%Improved
Train on diverse ReDWeb images rather than the indoor NYUDv2 dataMonocular relative depth model · DIW testWHDR: 27.70% -> 14.33%Improved
Fuse multi-scale features in the network (ablation)Proposed ResNet-50 model · DIW validationWHDR: 17.59% -> 15.74%Improved
Use the improved ranking loss with hard-pair selection instead of ordinary ranking loss (ablation)Proposed relative depth model · DIW validationWHDR: 16.67% -> 15.74%Improved
Use the improved ranking loss instead of ranking loss (ablation)Relative depth model trained on the NYUDv2 subset · NYUDv2WKDR: 35.8% -> 33.7%Improved
Increase the number of sampled point pairs per image (ablation)Proposed relative depth model · DIW validationWHDRImproved
Increase the number of ReDWeb training images (ablation)Proposed relative depth model · DIW validationWHDRImproved
Pre-train on ReDWeb before fine-tuning for metric depth estimationResNet-50 metric depth model · NYUDv2relImproved
Pre-train on ReDWeb before fine-tuning for semantic segmentationResNet-101 segmentation model · VOC 2012mIU: 69.67 -> 70.82Improved

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