Research map / 3D vision / Human pose estimation
3D Human Pose Estimation via Non-Causal Retentive Networks
Kaili Zheng, Feixiang Lu, Yihao Lv, Liangjun Zhang, Chenyi Guo, Ji Wu
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 changed | Where | Measured | Effect |
|---|---|---|---|
| train with a large chunk (243) and infer with a small chunk (27) instead of training at 27 (ablation) | NC-RetNet · Human3.6M (CPN), test T=27 | MPJPE (mm): 43.7 -> 42.1 | Improved |
| train with chunk 243 rather than 81, test at T=81 (ablation) | NC-RetNet · Human3.6M (CPN), test T=81 | MPJPE (mm): 41.9 -> 40.9 | Improved |
| add cross-chunk state to the RetNet model (ablation) | NC-RetNet · Human3.6M (CPN), train T=243, test T=27 | MPJPE (mm): 54.2 -> 42.1 | Improved |
| add cross-chunk state, test at the same size as training (T=243) (ablation) | NC-RetNet · Human3.6M (CPN), test T=243 | MPJPE (mm): 42.5 -> 40.4 | Improved |
| replace MixSTE temporal encoder with xPos (no state) versus NC-RetNet (ablation) | MixSTE · Human3.6M (CPN), train T=243, test T=27 | MPJPE (mm): 48.8 vs 42.1 | Hurt |
| apply NC-RetNet instead of STCFormer at small test chunk size | Human3.6M CPN · Human3.6M (CPN), T=27 vs 81 | MPJPE (mm): 42.1 (Ours T=27) vs 42.0 (STCFormer T=81) | Improved |
| NC-RetNet versus STCFormer at large test chunk (T=243) on PA-MPJPE | Human3.6M CPN · Human3.6M (CPN), T=243 | PA-MPJPE (mm): 32.5 vs 31.8 (STCFormer) | Hurt |
| use NC-RetNet with small chunk size on MPI-INF-3DHP | MPI-INF-3DHP · MPI-INF-3DHP, T=9 | MPJPE (mm): 28.2 -> 24.1 | Improved |
| use NC-RetNet with 2D ground truth keypoints at T=81 | Human3.6M GT · Human3.6M (2D ground truth input), T=81 | MPJPE (mm): 25.7 -> 22.4 | Improved |
| replace MixSTE temporal encoder with NC-RetNet, compute cost | MixSTE · Human3.6M (CPN), T=243 | FLOPs per frame (M): 572 -> 430 | Improved |
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.
- MixSTE — MixSTE: Seq2seq Mixed Spatio-Temporal Encoder for 3D Human Pose Estimation in Video (CVPR 2022) · better on 5 of 5 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
- Exploiting temporal information for 3D human pose estimation (ECCV 2018 · similarity 0.89)
- SRNet: Improving Generalization in 3D Human Pose Estimation with a Split-and-Recombine Approach (ECCV 2020 · similarity 0.89)
- From Human Pose Similarity Metric to 3D Human Pose Estimator: Temporal Propagating LSTM Networks (TPAMI 2022 · similarity 0.88)
- RepNet: Weakly Supervised Training of an Adversarial Reprojection Network for 3D Human Pose Estimation (CVPR 2019 · similarity 0.88)
- A Simple yet Effective Baseline for 3D Human Pose Estimation (ICCV 2017 · similarity 0.88)
- RePOSE: 3D Human Pose Estimation via Spatio-Temporal Depth Relational Consistency (ECCV 2024 · similarity 0.88)
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.
