Research map / Video / Video generation
From Image to Video: An Empirical Study of Diffusion Representations
Pedro Vélez, Luisa F. Polanía, Yi Yang, Chuhan Zhang, Rishabh Kabra, Anurag Arnab, Mehdi S. M. Sajjadi
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 |
|---|---|---|---|
| train the same WALT architecture for video instead of image generation | WALT 284M · Places365 | relative performance change: +0.6% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · ImageNet-1k | relative performance change: +1.8% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · iNaturalist 2018 | relative performance change: +11% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · Kinetics-400 | relative performance change: +8% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · Kinetics-700 | relative performance change: +12% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · Something-Something v2 | relative performance change: +42% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · Waymo box tracking | relative performance change: +23% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · RealEstate10K camera pose | relative performance change: +60% | Improved |
| train the same WALT architecture for video instead of image generation | WALT 284M · Perception Test point tracking | relative performance change: +68% | Improved |
| scale V-WALT from 284M to 1.9B parameters | V-WALT · most readout tasks | performance: 0.618 -> 0.728 | Improved |
| scale V-WALT from 284M to 1.9B parameters | V-WALT · Perception Test point tracking | average Jaccard: 0.756 -> 0.735 | Hurt |
| add small noise (timestep 200) to input latents before feature extraction (ablation) | V-WALT · most downstream tasks | relative performance | Improved |
Closest papers by meaning
- Emergent Temporal Correspondences from Video Diffusion Transformers (NeurIPS 2025 · similarity 0.90)
- Efficient Video Diffusion Models via Content-Frame Motion-Latent Decomposition (ICLR 2024 · similarity 0.90)
- Towards Precise Scaling Laws for Video Diffusion Transformers (CVPR 2025 · similarity 0.89)
- Dual Diffusion for Unified Image Generation and Understanding (CVPR 2025 · similarity 0.89)
- MeDM: Mediating Image Diffusion Models for Video-to-Video Translation with Temporal Correspondence Guidance (AAAI 2024 · similarity 0.89)
- LLM-grounded Video Diffusion Models (ICLR 2024 · similarity 0.89)
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.
