Research map / Video / Video segmentation and tracking
Learning Temporal Transformations from Time-Lapse Videos
Yipin Zhou, Tamara L. Berg
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 gradient-domain MSE loss (ablation) | pairwise p_mse+adv · Time-lapse test set (Pairwise generation) | PSNR: 17.0409 -> 17.0660 | Improved |
| add gradient-domain MSE loss (ablation) | pairwise p_mse+adv · Time-lapse test set (Pairwise generation) | SSIM: 0.5576 -> 0.5720 | Improved |
| add reconstruction pre-training then fine-tuning (ablation) | pairwise p_g_mse+adv · Time-lapse test set (Pairwise generation) | SSIM: 0.5720 -> 0.6036 | Improved |
| add reconstruction pre-training then fine-tuning (ablation) | two-stack p_g_mse+adv · Time-lapse test set (Two stack generation) | PSNR: 17.9157 -> 18.6812 | Improved |
| add reconstruction pre-training then fine-tuning (ablation) | recurrent p_g_mse+adv · Time-lapse test set (Recurrent generation) | PSNR: 17.2951 -> 18.3357 | Improved |
| add gradient-domain MSE loss (ablation) | two-stack p_mse+adv · Time-lapse test set (Two stack generation) | SSIM: 0.5970 -> 0.6122 | Improved |
| add gradient-domain MSE loss (ablation) | recurrent p_mse+adv · Time-lapse test set (Recurrent generation) | MSE: 0.0211 -> 0.0214 | Hurt |
| add adversarial loss to pixel MSE baseline (ablation) | pairwise p_mse · Human preference (Pairwise generation) | fraction chosen as best: 0.1490 -> 0.2360 | Improved |
| add pre-training plus fine-tuning (ablation) | pairwise p_g_mse+adv · Human preference (Pairwise generation) | fraction chosen as best: 0.2765 -> 0.3385 | Improved |
| add pre-training plus fine-tuning (ablation) | two-stack p_g_mse+adv · Human preference (Two stack generation) | fraction chosen as best: 0.3050 -> 0.3790 | Improved |
| add gradient loss (ablation) | two-stack p_mse+adv · Human preference (Two stack generation) | fraction chosen as best: 0.3160 -> 0.3050 | Hurt |
| add pre-training plus fine-tuning (ablation) | recurrent p_g_mse+adv · Human preference (Recurrent generation) | fraction chosen as best: 0.3130 -> 0.3815 | Improved |
Closest papers by meaning
- Generating the Future With Adversarial Transformers (CVPR 2017 · similarity 0.90)
- Learning Temporal Dynamics From Cycles in Narrated Video (ICCV 2021 · similarity 0.88)
- ChronoMagic-Bench: A Benchmark for Metamorphic Evaluation of Text-to-Time-lapse Video Generation (NeurIPS 2024 · similarity 0.87)
- Dreamweaver: Learning Compositional World Models from Pixels (ICLR 2025 · similarity 0.87)
- Generative Perception of Shape and Material from Differential Motion (NeurIPS 2025 · similarity 0.86)
- PlaySlot: Learning Inverse Latent Dynamics for Controllable Object-Centric Video Prediction and Planning (ICML 2025 · similarity 0.86)
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.
