Research map / Video / Video generation
DisCoRD: Discrete Tokens to Continuous Motion via Rectified Flow Decoding
Jungbin Cho, Junwan Kim, Jisoo Kim, Minseo Kim, Mingu Kang, Sungeun Hong, Tae-Hyun Oh, Youngjae Yu
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 |
|---|---|---|---|
| replace feed-forward decoder with DisCoRD rectified flow decoder | MoMask · HumanML3D reconstruction | sJPE: 0.512 -> 0.385 | Improved |
| replace decoder with DisCoRD | MoMask · HumanML3D text-to-motion | FID: 0.045 -> 0.032 | Improved |
| replace decoder with DisCoRD | T2M-GPT · HumanML3D text-to-motion | R-Precision Top 3: 0.775 -> 0.760 | Hurt |
| replace decoder with DisCoRD | MoMask · HumanML3D reconstruction | MPJPE: 29.5 -> 33.3 | Hurt |
| replace decoder with DisCoRD on TM2D | TM2D · AIST++ music-to-dance | FID_g: 20.09 -> 88.74 | Hurt |
| use DisCoRD conditioned decoding instead of rectified flow post refinement (ablation) | MoMask · HumanML3D text-to-motion | FID: 0.035 -> 0.032 | Improved |
| use repeat-stack-project condition projection instead of up-convolution (ablation) | DisCoRD · HumanML3D text-to-motion | FID: 0.039 -> 0.032 | Improved |
| add attention to U-Net backbone (ablation) | DisCoRD · HumanML3D text-to-motion | FID: 0.032 -> 0.043 | Hurt |
| train on sliding windows instead of full motion sequences (ablation) | DisCoRD · HumanML3D text-to-motion | FID: 0.038 -> 0.032 | Improved |
| increase residual quantization level of the tokenizer (ablation) | MoMask + DisCoRD · HumanML3D text-to-motion | R-Precision gain: RP@1 -2.8% at R0 -> +0.6% at R5 | Improved |
| use rectified flow with ODE sampler instead of diffusion DDIM at 2 steps (ablation) | DisCoRD decoder · HumanML3D decoding | FID: 0.055 -> 0.034 (2 steps) | Improved |
| use sJPE instead of MPJPE for human-perceived naturalness | evaluation metric · user study correlation | Pearson correlation with human score: 0.181 -> 0.483 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- BAMM — BAMM: Bidirectional Autoregressive Motion Model (ECCV 2024) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Seamless Human Motion Composition with Blended Positional Encodings (CVPR 2024 · similarity 0.88)
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.88)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.87)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.87)
- TokenMotion: Decoupled Motion Control via Token Disentanglement for Human-centric Video Generation (CVPR 2025 · similarity 0.87)
- Scaling Large Motion Models with Million-Level Human Motions (ICML 2025 · similarity 0.87)
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.
