M$^3$GPT: An Advanced Multimodal, Multitask Framework for Motion Comprehension and Generation

This paper presents M$^3$GPT, an advanced $\textbf{M}$ultimodal, $\textbf{M}$ultitask framework for $\textbf{M}$otion comprehension and generation. M$^3$GPT operates on three fundamental principles. The first focuses on creating a unified representation space for various motion-relevant modalities. We employ discrete vector quantization for multimodal conditional signals, such as text, music and motion/dance, enabling seamless integration into a large language model (LLM) with a single vocabulary. The second involves modeling motion generation directly in the raw motion space. This strategy circumvents the information loss associated with a discrete tokenizer, resulting in more detailed and comprehensive motion generation. Third, M$^3$GPT learns to model the connections and synergies among various motion-relevant tasks. Text, the most familiar and well-understood modality for LLMs, is utilized as a bridge to establish connections between different motion tasks, facilitating mutual reinforcement. To our knowledge, M$^3$GPT is the first model capable of comprehending and generating motions based on multiple signals. Extensive experiments highlight M$^3$GPT's superior performance across various motion-relevant tasks and its powerful zero-shot generalization capabilities for extremely challenging tasks. Project page: \url{https://github.com/luomingshuang/M3GPT}.

Paper

References (63)

Scroll for more · 38 remaining

Similar papers

Peer review

Reviewer 2Gpw5/10 · confidence 4/52024-07-11

Summary

This paper introduces M3GPT, an advanced Multimodal, Multitask framework for motion comprehension and generation. M3GPT operates based on three fundamental principles. Firstly, it aims to create a unified representation space for various motion-relevant modalities. We employ discrete vector quantization for multimodal control and generation signals, such as text, music, and motion/dance, enabling seamless integration into a large language model (LLM) with a single vocabulary. Secondly, it involves modeling directly in the raw motion space. This strategy circumvents the information loss associated with discrete tokenizers, resulting in more detailed and comprehensive model generation. Thirdly, M3GPT learns to model the connections and synergies among various motion-relevant tasks. Text, being the most familiar and well-understood modality for LLMs, is utilized as a bridge to establish connections between different motion tasks, facilitating mutual reinforcement.

Strengths

1. M3GPT is equipped with multimodal tokenizers capable of compressing raw multimodal data, including motion, music, and text, into a sequence of discrete semantic tokens. 2. The paper jointly trains LLM and motion de-tokenizer, optimizing LLM in both discrete semantic space and raw continuous motion space. 3. This Paper is well writing and the video results in Supp. are well presented.

Weaknesses

1. In Table 3, some comparation results are missed. For text to motion generation task, you could compare your method with 1. MoMask: Generative Masked Modeling of 3D Human Motions; 2. Plan, posture, and go: Towards Open-World Text-to-Motion Generation; and 3. Real-time Controllable Motion Generation via Latent Consistency Model. If it’s not suitable for the comparations, you could explain the reason. 2. The parameterizations (dimensions) of the variable in Section 3 are all missing. 3. In line 167, T5 is utilized as the language model backbone. While most of the previous model utilize CLIP or BERT, you should include ablation study about the text encoder.

Questions

1. Is it just a matter of migrating multimodal model to the motion domain? 2. What new issues have been encountered in the process? 3. What is your innovation in solving these problems?

Rating

5

Confidence

4

Soundness

3

Presentation

2

Contribution

2

Limitations

The author has addressed the limitations. This paper focuses on modeling human body movements, excluding hands and faces details.

Authorsrebuttal2024-08-12

Any questions about the rebuttal

Dear Reviewer 2Gpw: As the rebuttal period is ending soon, please let us know whether your concerns have been addressed or not, and if there are any further questions.

Reviewer KMVC6/10 · confidence 4/52024-07-11

Summary

The paper presents M3GPT, which creates a unified representation space for different motion-relevant modalities, including text, music, and motion/dance. The framework employs discrete vector quantization for these modalities, enabling integration into a large language model (LLM) with a single vocabulary. The model operates directly in the raw motion space to avoid information loss associated with discrete tokenizers. Extensive experiments highlight M3GPT’s superior performance across various motion tasks and its zero-shot generalization capabilities.

Strengths

- The paper introduces a novel approach to integrate multiple modalities (text, music, motion) within a single framework, addressing a significant gap in existing research. The unified representation space and the strategy of using text as a bridge to connect different modalities are theoretically sound and contribute to the field of multimodal learning. - M3GPT handles six core tasks related to motion comprehension and generation, demonstrating versatility and robustness. The experiments show that M3GPT achieves competitive performance across multiple tasks. - The model's ability to perform well in zero-shot scenarios, such as long-term dance generation and music-text conditioned dance generation, is particularly impressive and valuable for practical applications.

Weaknesses

- The proposed framework is complex and may require significant computational resources to implement and extend. - While the experimental results are promising, the paper could benefit from a more detailed discussion of the evaluation metrics and benchmarks used to assess performance. A more extensive comparison with a broader range of baselines could strengthen the argument for its superiority. Also, the paper would benefit from a more in-depth discussion on the potential real-world applications and limitations of M3GPT.

Questions

- Could you provide more details on the training process, including the computational resources required and the time taken to train M3GPT? - Have you conducted any ablation studies to understand the contribution of each component of M3GPT to the overall performance? If so, could you include those results? - Can you provide more examples or case studies where M3GPT has been applied to real-world scenarios? What challenges were encountered?

Rating

6

Confidence

4

Soundness

3

Presentation

3

Contribution

3

Limitations

Limitations have been discussed.

Authorsrebuttal2024-08-12

Any questions about the rebuttal

Dear Reviewer KMVC: As the rebuttal period is ending soon, please let us know whether your concerns have been addressed or not, and if there are any further questions.

Reviewer Wkeh5/10 · confidence 5/52024-07-11

Summary

This paper presents an advanced Multimodal, Multitask framework for Motion comprehension and generation. It aims to create a unified representation space for various motion-relevant modalities and model the connections and synergies among different motion tasks. M3GPT consists of multimodal tokenizers and a motion-aware language model, and is trained through a multimodal pre-training + instruction-tuning pipeline. The model is evaluated on multiple motion-relevant tasks and datasets, and shows competitive performance and strong zero-shot generalization capabilities.

Strengths

(1) Unifies multiple motion-related tasks in a single model, achieving bidirectional alignment. (2) Achieves competitive performance across multiple tasks and datasets and strong zero-shot generalization capabilities. (3) Explores motion related music2text and text2dance as auxiliary tasks.

Weaknesses

(1) For supplementary videos, it's hard to judge the quality of music-motion bidirectional generations without combining them in one video; also, long-term dance generation on aist++ dataset is only a little longer than 5s. (2) The method of Joint optimization of LLM and motion de-tokenizer lacks important details. In line 194-195, I wonder how the gradient from de-tokenizer backprogate to LLM if author use softmax to choose top-1 token. (3) For motion prediction and In-between tasks, feature-level metrics like FID are not enough, pose-level metrics like MPJPE are needed for comparison.

Questions

(1) Did authors try different LLM architecture or different size of T5, what is the conclusion and the training time for T5 base? (2) Did you evalute two auxiliary tasks music2text or text2dance quantitatively or qualitatively? (3) For zero-shot tasks, can they combined together for more application? For instance, generating a long dance while we specify text prompt for one segment in the middle of sequence. Also, how about zero-shot text2music? (4) For tab.4, what is the result of Instruction-tuned only on AIST++ or FineDance of which the setting is similar to the last row in tab.3. (5) How did you implement motion in-between in pretrained stage use the equation(4)? More specifically, whether your model is conditioned on future target frame, if so, the equation is not precise enough. (6) Is there any special token for motion and music in the unified token vocabulary, like start/end of motion/music token? If not, how to ensure the pretrained model generate desired modality? For example, I wanna directly generate dance conditioned on music, but actually the model generate text token. (7) The statistics of music2text and text2dance paired dataset are missing. I would like to raise the score if authors can well address my concerns questions.

Rating

5

Confidence

5

Soundness

3

Presentation

3

Contribution

3

Limitations

N.A

Authorsrebuttal2024-08-12

Any questions about the rebuttal

Dear Reviewer Wkeh: As the rebuttal period is ending soon, please let us know whether your concerns have been addressed or not, and if there are any further questions.

Reviewer ZwnC5/10 · confidence 4/52024-07-13

Summary

In this paper, the authors introduce $M^3$GPT, a multimodal multitask framework designed for both motion comprehension and generation. Utilizing discrete vector quantization, $M^3$GPT establishes a discrete semantic representation space for various modalities. To avoid information loss during discrete de-tokenization, $M^3GPT$ jointly trains the large language model (LLM) and motion de-tokenizer, thereby optimizing the LLM within both the discrete semantic space and the continuous raw motion space. Furthermore, the authors construct paired text descriptions for music and devise two supplementary tasks music-to-text and text-to-dance, which facilitate the alignment of music and dance within the text embedding space. The efficacy of this approach is demonstrated through experiments across a range of motion-related tasks, and showcases zero-shot generalization capabilities for highly challenging tasks.

Strengths

1. The motivation behind investigating the integration of text, music, and motion for motion comprehension and generation is convincing. 2. The proposed joint optimization of the large language model (LLM) and motion de-tokenizer, along with the synergy learning strategy incorporating auxiliary tasks, demonstrates potential in enhancing multimodal motion generation.

Weaknesses

1. The effectiveness of combining text, music, and motion modalities in the $M^3$GPT framework is not thoroughly evaluated. Although the authors present the $M^3$GPT capabilities in handling zero-shot tasks. However, there is still doubt about how can text-to-motion or music-to-dance tasks benefit from the multimodal framework. 2. Some key experiments are missing. a. In Table 3, the authors evaluate proposed $M^3$GPT on text-to-motion dataset Motion-X on 4 tasks. However, MotionGPT[1] can also perform the tasks while the comparison between MotionGPT[1] is missing. b. Besides motion prediction lacks comparison with models like T2M-GPT[2], since the auto-regressive models are better at prediction. c. Motion inbetweening lacks comparison with the state-of-the-art models like MoMask[3] capable of temporal inpainting which can handle motion inbetweening tasks. 3. The qualitative comparison of generated motions by the proposed $M^3$GPT framework from text or music is absent. The inclusion of detailed qualitative results is crucial for providing a comprehensive understanding of the naturalness and overall realism of the generated gestures. Minor: 1. In Figure 2, there is an error in drawing graphics in the Motion Codebook part. [1] Jiang, Biao, et al. "Motiongpt: Human motion as a foreign language." Advances in Neural Information Processing Systems 36 (2024). [2] Zhang, Jianrong, et al. "Generating human motion from textual descriptions with discrete representations." Proceedings of the IEEE/CVF conference on computer vision and pattern recognition. 2023. [3] Guo, Chuan, et al. "Momask: Generative masked modeling of 3d human motions." Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition. 2024.

Questions

1. I am curious about the long-duration dance generation mentioned in Page 6, especially if the LLM can maintain coherence and consistency in generating long sequences or this long-duration generation is achieved by the consistency during de-tokenization. 2. In Table 4, the authors evaluate music-to-dance on two datasets AIST++ and FineDance, $M^3$GPT achieves state-of-the-art performance on AIST++, but the performance on FineDance is not that competitive. Could the authors provide more insights into the reasons behind this performance gap between the two datasets? 3. Have authors evaluate different sizes of the $M^3$GPT model? Given the added modalities and tasks, it would be interesting to understand how the model scales with respect to performance and computational requirements.

Rating

5

Confidence

4

Soundness

3

Presentation

3

Contribution

2

Limitations

The authors have discussed limitations.

Authorsrebuttal2024-08-12

Any questions about the rebuttal

Dear Reviewer ZwnC: As the rebuttal period is ending soon, please let us know whether your concerns have been addressed or not, and if there are any further questions.

Authorsrebuttal2024-08-12

Reply to ethics review

Thanks for your review. We appreciate your feedback. We will increase the discussion of the broader impacts and potential ethical issues concerning our proposed $M^3$GPT in final version. **Ethical Issues.** There are several ethical issues. On one hand, if the motion data is derived from real individuals, there are concerns about the privacy of the data used for training and potential misuse of generated motions that resemble real people. On the other hand, the ability to generate realistic human motions could be misused to create misleading or deceptive content, such as in deepfake videos or fraudulent activities. **Board Impacts.** The proposed framework boosts the performance of motion comprehension and generation, making it more practical in AR/VR, animation creation, digital human design, etc. Meanwhile, the higher performance may raise  potential risks, such as privacy breaches and the misuse of deepfake human motions.  Therefore the usage of motion data and the proposed framework should be regulated.

Program Chairsdecision2024-09-25

Decision

Accept (poster)

© 2026 NYSGPT2525 LLC