# Distill Knowledge from NRSfM for Weakly Supervised 3D Pose Learning

arXiv: https://arxiv.org/abs/1908.06377 · HTML: https://mlsanity.com/paper/1908.06377/

## Key facts

- Cited by 13 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 18 Aug 2019.

## Built on by (most cited first, 13 of 13)

- [DanceTrack: Multi-Object Tracking in Uniform Appearance and Diverse Motion](https://mlsanity.com/paper/2111.14690/) (2021)
- [Online Knowledge Distillation for Efficient Pose Estimation](https://mlsanity.com/paper/2108.02092/) (2021)
- [ElePose: Unsupervised 3D Human Pose Estimation by Predicting Camera Elevation and Learning Normalizing Flows on 2D Poses](https://mlsanity.com/paper/2112.07088/) (2021)
- [Estimating Egocentric 3D Human Pose in the Wild with External Weak Supervision](https://mlsanity.com/paper/2201.07929/) (2022)
- [KAMA: 3D Keypoint Aware Body Mesh Articulation](https://mlsanity.com/paper/2104.13502/) (2021)
- [Weakly-supervised 3D Human Pose Estimation with Cross-view U-shaped Graph Convolutional Network](https://mlsanity.com/paper/2105.10882/) (2021)
- [On Triangulation as a Form of Self-Supervision for 3D Human Pose Estimation](https://mlsanity.com/paper/2203.15865/) (2022)
- [MHR-Net: Multiple-Hypothesis Reconstruction of Non-Rigid Shapes from 2D Views](https://mlsanity.com/paper/2207.09086/) (2022)
- [TriPose: A Weakly-Supervised 3D Human Pose Estimation via Triangulation from Video](https://mlsanity.com/paper/2105.06599/) (2021)
- [End-to-End Learning of Multi-category 3D Pose and Shape Estimation](https://mlsanity.com/paper/2112.10196/) (2021)
- ["Teaching Independent Parts Separately" (TIPSy-GAN) : Improving Accuracy and Stability in Unsupervised Adversarial 2D to 3D Pose Estimation](https://mlsanity.com/paper/2205.05980/) (2022)
- [Mirror-Aware Neural Humans](https://mlsanity.com/paper/2309.04750/) (2023)
- [LaxMotion: Rethinking Supervision Granularity for 3D Human Motion Generation](https://mlsanity.com/paper/2511.11368/) (2025)

## Abstract (arXiv)

We propose to learn a 3D pose estimator by distilling knowledge from Non-Rigid Structure from Motion (NRSfM). Our method uses solely 2D landmark annotations. No 3D data, multi-view/temporal footage, or object specific prior is required. This alleviates the data bottleneck, which is one of the major concern for supervised methods. The challenge for using NRSfM as teacher is that they often make poor depth reconstruction when the 2D projections have strong ambiguity. Directly using those wrong depth as hard target would negatively impact the student. Instead, we propose a novel loss that ties depth prediction to the cost function used in NRSfM. This gives the student pose estimator freedom to reduce depth error by associating with image features. Validated on H3.6M dataset, our learned 3D pose estimation network achieves more accurate reconstruction compared to NRSfM methods. It also outperforms other weakly supervised methods, in spite of using significantly less supervision.