Research map / Video / Video generation
Quantized GAN for Complex Music Generation from Dance Videos
Ye Zhu, Kyle Olszewski, Yu Wu, Panos Achlioptas, Menglei Chai, Yan Yan, Sergey Tulyakov
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 |
|---|---|---|---|
| add motion input (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Genre accuracy: 15.1 -> 26.7 | Improved |
| add visual input (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Genre accuracy: 17.4 -> 26.7 | Improved |
| add motion input (ablation) | D2M-GAN High-level · AIST++ test (ablation) | Beats Coverage: 83.5 -> 88.2 | Improved |
| add mel-spectrogram loss (ablation) | D2M-GAN High-level · AIST++ test (ablation) | Beats Coverage: 77.6 -> 88.2 | Improved |
| add mel-spectrogram loss (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Beats Hit: 77.1 -> 91.7 | Improved |
| add feature matching loss (ablation) | D2M-GAN High-level · AIST++ test (ablation) | Beats Coverage: 85.3 -> 88.2 | Improved |
| add waveform L1 loss (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Beats Coverage: 89.4 -> 92.3 | Improved |
| use 3-layer multi-scale discriminator instead of 1 layer (ablation) | D2M-GAN High-level · AIST++ test (ablation) | Genre accuracy: 9.3 -> 24.4 | Improved |
| add output scaling in the generator (ablation) | D2M-GAN High-level · AIST++ test (ablation) | Beats Coverage: 72.9 -> 88.2 | Improved |
| add temporal reshape of VQ features for discriminators (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Beats Hit: 71.8 -> 91.7 | Improved |
| fine-tune JukeBox decoder (ablation) | D2M-GAN Low-level · AIST++ test (ablation) | Beats Hit: 91.2 -> 91.7 | No clear effect |
| lengthen training sequences from 2s to 3s (ablation) | D2M-GAN High-level · AIST++ test | Beats Hit: 84.7 -> 85.3 | Mixed |
Closest papers by meaning
- MuseGAN: Multi-track Sequential Generative Adversarial Networks for Symbolic Music Generation and Accompaniment (AAAI 2018 · similarity 0.91)
- MNET++: Music-Driven Pluralistic Dancing Toward Multiple Dance Genre Synthesis (TPAMI 2023 · similarity 0.90)
- Symbolic Music Generation with Transformer-GANs (AAAI 2021 · similarity 0.90)
- DanceAnyWay: Synthesizing Beat-Guided 3D Dances with Randomized Temporal Contrastive Learning (AAAI 2024 · similarity 0.90)
- OmniDance: Multimodal Driven Dance Video Generation with Large-scale Internet Data (ECCV 2026 · similarity 0.90)
- V2Meow: Meowing to the Visual Beat via Video-to-Music Generation (AAAI 2024 · similarity 0.88)
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.
