|
|
--- |
|
|
license: mit |
|
|
tags: |
|
|
- smpl |
|
|
- human-pose-and-shape-estimation |
|
|
- human-mesh-recovery |
|
|
- inverse-kinematics |
|
|
pipeline_tag: keypoint-detection |
|
|
--- |
|
|
|
|
|
# Learnable SMPLify: A Neural Solution for Optimization-Free Human Pose Inverse Kinematics |
|
|
|
|
|
#### <p align="center">[arXiv Paper](https://arxiv.org/abs/2508.13562) | [Code](https://github.com/Charrrrrlie/Learnable-SMPLify)</p> |
|
|
|
|
|
## Abstract |
|
|
In 3D human pose and shape estimation, SMPLify remains a robust baseline that solves inverse kinematics (IK) through iterative optimization. However, its high computational cost limits its practicality. Recent advances across domains have shown that replacing iterative optimization with data-driven neural networks can achieve significant runtime improvements without sacrificing accuracy. Motivated by this trend, we propose Learnable SMPLify, a neural framework that replaces the iterative fitting process in SMPLify with a single-pass regression model. The design of our framework targets two core challenges in neural IK: data construction and generalization. To enable effective training, we propose a temporal sampling strategy that constructs initialization-target pairs from sequential frames. To improve generalization across diverse motions and unseen poses, we propose a human-centric normalization scheme and residual learning to narrow the solution space. Learnable SMPLify supports both sequential inference and plug-in post-processing to refine existing image-based estimators. Extensive experiments demonstrate that our method establishes itself as a practical and simple baseline: it achieves nearly 200x faster runtime compared to SMPLify, generalizes well to unseen 3DPW and RICH, and operates in a model-agnostic manner when used as a plug-in tool on LucidAction. |
|
|
|
|
|
--- |
|
|
|
|
|
``TL;DR`` Given X_{t-s} and X_{t} 3D keypoints, |
|
|
calculate residual SMPL parameters from t-s to t. |
|
|
|
|
|
## Sample Usage |
|
|
To run sequential inference using the trained model, navigate to the cloned repository and execute the following command: |
|
|
|
|
|
```bash |
|
|
python inference.py <PATH_TO_CHECKPOINT> (<DATASET_NAME> <SAMPLE_RATIO>) |
|
|
``` |
|
|
For detailed installation and data preparation, as well as instructions for training and evaluation, please refer to the [GitHub repository](https://github.com/Charrrrrlie/Learnable-SMPLify). |
|
|
|
|
|
## Citation |
|
|
If you find this work useful in your research, please consider citing: |
|
|
|
|
|
``` |
|
|
@misc{LearnableSMPLify, |
|
|
title={Learnable SMPLify: A Neural Solution for Optimization-Free Human Pose Inverse Kinematics}, |
|
|
author={Yuchen, Yang and Linfeng, Dong and Wei, Wang and Zhihang, Zhong and Xiao, Sun}, |
|
|
year={2025}, |
|
|
eprint={2508.13562}, |
|
|
archivePrefix={arXiv}, |
|
|
primaryClass={cs.CV} |
|
|
} |
|
|
``` |
