AI Paper Atlas

Research map / 3D vision / NeRF and radiance fields

Motion-Oriented Compositional Neural Radiance Fields for Monocular Dynamic Human Modeling

Jaehyeok Kim, Dongyoon Wee, Dan Xu

ECCV 2024 · 3D vision › NeRF and radiance fields · approach: Sparse-input NeRF regularization · 3 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
add pose-embedded implicit features to the non-rigid branch (ablation)MoCo-NeRF · ZJU-MoCap (single-subject)PSNR: 30.64 -> 31.06Improved
add radiance residual (non-rigid) branch on top of the rigid branch (ablation)MoCo-NeRF · ZJU-MoCap (single-subject)LPIPS x1000: 29.93 -> 28.83Improved
use cross-attention to generate the pose-embedded feature (ablation)MoCo-NeRF · ZJU-MoCap (single-subject)LPIPS x1000: 30.66 -> 28.83Improved
share frozen canonical features with the residual branch instead of unfrozen or no sharing (ablation)MoCo-NeRF · ZJU-MoCap (single-subject)LPIPS x1000: 30.17 (no sharing), 30.71 (unfrozen) -> 28.83Improved
add identity codes for multi-subject training (ablation)multi-subject MoCo-NeRF · ZJU-MoCap (multi-subject)PSNR: 28.63 -> 30.70Improved
add N_s local MHEs for multi-subject training (ablation)multi-subject MoCo-NeRF · ZJU-MoCap (multi-subject)PSNR: 28.63 -> 30.86Improved
add global MHE shared across subjects (ablation)multi-subject MoCo-NeRF · ZJU-MoCap (multi-subject)LPIPS x1000: 28.49 -> 27.69Improved
split one MHE's features between rigid and non-rigid branches instead of two separate MHEs (ablation)MoCo-NeRF · training and rendering speedspeed overhead: avoids x1.14 slowdownImproved
replace HumanNeRF canonical and non-rigid MLPs with MHEs (ablation)HumanNeRF · ZJU-MoCapqualitative renderingHurt
train longerGauHuman and Instant-NVR · ZJU-MoCapnon-rigid motion qualityHurt
MoCo-NeRF vs HumanNeRF on MonoCap perceptual metricMoCo-NeRF · MonoCap (single-subject)LPIPS x1000: 12.93 (HumanNeRF) vs 13.02No clear effect
use MoCo-NeRF instead of HumanNeRF for training efficiency (6 subjects)MoCo-NeRF · ZJU-MoCap (6 subjects)train time (hours): 233.3 -> 2.1Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 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