Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration

TL;DR

A novel computationally efficient encoder-decoder model with uncertainty consideration is proposed, which could learn proper characteristics for future frames by a dynamic function, which could produce motion sequences with much better quality that avoids the intractable shaking artefacts.

Abstract

Human motion prediction is consisting in forecasting future body poses from historically observed sequences. It is a longstanding challenge due to motion's complex dynamics and uncertainty. Existing methods focus on building up complicated neural networks to model the motion dynamics. The predicted results are required to be strictly similar to the training samples with L2 loss in current training pipeline. However, little attention has been paid to the uncertainty property which is crucial to the prediction task. We argue that the recorded motion in training data could be an observation of possible future, rather than a predetermined result. In addition, existing works calculate the predicted error on each future frame equally during training, while recent work indicated that different frames could play different roles. In this work, a novel computationally efficient encoder-decoder model with uncertainty consideration is proposed, which could learn proper characteristics for future frames by a dynamic function. Experimental results on benchmark datasets demonstrate that our uncertainty consideration approach has obvious advantages both in quantity and quality. Moreover, the proposed method could produce motion sequences with much better quality that avoids the intractable shaking artefacts. We believe our work could provide a novel perspective to consider the uncertainty quality for the general motion prediction task and encourage the studies in this field. The code will be available in https://github.com/Motionpre/Adaptive-Salient-Loss-SAGGB.

Figures (5)

• Figure 1: The uncertainty characteristic of human motion. For a certain motion clip, the recorded future motion should be an observation of possible future, rather than a predetermined result. The future motion could vary even with similar historical motion.
• Figure 2: Training pipeline of existing prediction approaches. A given piece of training data sample would be divided into past pose sequences $X_{obs }$ and future pose sequences $X_{pre }$. The predicted result $\hat{X}_{pre }$ is required to be strictly similar to the training data $X_{pre }$, where the uncertainty of future motion has been ignored.
• Figure 3: Overview of proposed model for human motion prediction with SAGGB. In the encoder, SAGGB leverage self attention mechanism to generate sample related graph to extract spatial information. In the decoder, we use lightweight CNNs and MLP to predict.
• Figure 4: Comparison of Qualitative Evaluation Results. Joint trajectory is recorded from 400ms to 1000ms. The shaking problems have occurred in comparison method's prediction results, which could be caused by over-fitting.
• Figure 5: Ablation on value of $\lambda$ and $\omega$ on H3.6M