AI Paper Atlas

Research map / 3D vision / Human pose estimation

Cross-Attention of Disentangled Modalities for 3D Human Mesh Recovery with Transformers

Junhyeong Cho, Kim Youwang, Tae-Hyun Oh

ECCV 2022 · 3D vision › Human pose estimation · approach: Human mesh recovery from images · 123 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 attention masking of non-adjacent vertices (ablation)FastMETRO-S-R50 · Human3.6MMPJPE: 58.0 -> 55.7Improved
use graph convolutions instead of attention masking (ablation)FastMETRO-S-R50 · Human3.6MMPJPE: 55.7 -> 56.6No clear effect
mask attention in only half of the heads (ablation)FastMETRO-S-R50 · Human3.6MMPJPE: 55.7 -> 55.8No clear effect
use precomputed upsampling matrix instead of learnable linear layers (ablation)FastMETRO-S-R50 · Human3.6MMPJPE: 58.1 -> 55.7Improved
add progressive dimensionality reduction (ablation)FastMETRO-S-R50 · Human3.6Mparams with similar accuracy: 39.5M -> 32.7MImproved
use fixed instead of learnable positional encodings (ablation)FastMETRO-S-R50 · Human3.6MPA-MPJPE: 39.7 -> 39.4Improved
add camera token in the transformer decoder (ablation)FastMETRO-S-R50 · Human3.6MPA-MPJPE: 39.5 -> 39.4No clear effect
predict camera from estimated mesh (ablation)FastMETRO-S-R50 · Human3.6MMPJPE: 55.7 -> 58.0Hurt
full FastMETRO design vs naive encoder-decoder baseline (ablation)transformer encoder-decoder · Human3.6MPA-MPJPE: 41.8 -> 39.4Improved
scale transformer from S to L variantFastMETRO R50 · Human3.6MPA-MPJPE: 39.4 -> 37.3Improved
swap ResNet-50 for HRNet-W64 backboneFastMETRO-L · Human3.6MPA-MPJPE: 37.3 -> 33.7Improved
train longer, 60 to 200 epochsMETRO-H64 · Human3.6MPA-MPJPE: 38.0 -> 36.7Improved

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