Research map / 3D vision / Human pose estimation
SeqHAND: RGB-Sequence-Based 3D Hand Pose and Shape Estimation
John Yang, Hyung Jin Chang, Seungeui Lee, Nojun Kwak
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 |
|---|---|---|---|
| add a ConvLSTM layer to the ResNet-50 encoder (ablation) | ResNet50 encoder · SeqHAND validation | 3D error (mm): 7.85 -> 7.18 | Improved |
| add a plain LSTM layer to the ResNet-50 encoder (ablation) | ResNet50 encoder · SeqHAND validation | 3D error (mm): 7.85 -> 9.63 | Hurt |
| replace encoder with I3D 3D convolution (ablation) | ResNet50 encoder · SeqHAND validation | 3D error (mm): 7.85 -> 9.24 | Hurt |
| replace encoder with MFNet motion-feature network (ablation) | ResNet50 encoder · SeqHAND validation | 3D error (mm): 7.85 -> 10.54 | Hurt |
| use ResNet101 instead of ResNet50 backbone (ablation) | encoder baseline · SeqHAND validation | 3D error (mm): 7.85 -> 7.54 | Improved |
| detach ConvLSTM from real-domain finetuning (ablation) | SeqHAND-Net · EgoDexter (ED) | avg 3D error (mm): 40.94 -> 17.16 | Improved |
| detach ConvLSTM from real-domain finetuning (ablation) | SeqHAND-Net · Dexter+Object (DO) | avg 3D error (mm): 29.41 -> 18.12 | Improved |
| detach ConvLSTM from real-domain finetuning (ablation) | SeqHAND-Net · STB | avg 3D error (mm): 9.82 -> 9.87 | No clear effect |
| add SeqHAND sequential pretraining with ConvLSTM, then finetune with ConvLSTM also finetuned (vs non-sequential synthetic pretraining) (ablation) | encoder · EgoDexter (ED) | avg 3D error (mm): 49.18 -> 40.94 | Improved |
| add SeqHAND sequential pretraining with ConvLSTM, then finetune with ConvLSTM also finetuned (vs non-sequential synthetic pretraining) (ablation) | encoder · Dexter+Object (DO) | avg 3D error (mm): 33.12 -> 29.41 | Improved |
| add SeqHAND sequential pretraining with ConvLSTM, finetuned jointly (vs non-sequential synthetic pretraining) (ablation) | encoder · STB | avg 3D error (mm): 9.80 -> 9.82 | No clear effect |
| finetune on FreiHand+STB after synthetic pretraining (ablation) | encoder (no ConvLSTM) · Dexter+Object (DO) | avg 3D error (mm): 100.84 -> 33.12 | Improved |
Closest papers by meaning
- Pushing the Envelope for RGB-Based Dense 3D Hand Pose Estimation via Neural Rendering (CVPR 2019 · similarity 0.91)
- InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image (ECCV 2020 · similarity 0.91)
- FreiHAND: A Dataset for Markerless Capture of Hand Pose and Shape From Single RGB Images (ICCV 2019 · similarity 0.91)
- First-Person Hand Action Benchmark With RGB-D Videos and 3D Hand Pose Annotations (CVPR 2018 · similarity 0.91)
- Keypoint Fusion for RGB-D Based 3D Hand Pose Estimation (AAAI 2024 · similarity 0.90)
- Analyzing the Synthetic-to-Real Domain Gap in 3D Hand Pose Estimation (CVPR 2025 · similarity 0.89)
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.
