Research map / Video / Video segmentation and tracking
Dual Motion GAN for Future-Flow Embedded Video Prediction
Xiaodan Liang, Lisa Lee, Wei Dai, Eric P. Xing
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 the probabilistic motion encoder (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00262 -> 0.00241 | Improved |
| add the probabilistic motion encoder (ablation) | dual motion GAN · YouTube Clips (trained on KITTI) | MSE: 0.00583 -> 0.00558 | Improved |
| add adversarial discriminators to the frame+flow model (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00269 -> 0.00241 | Improved |
| add frame discriminator (GAN) to frame generator (ablation) | frame generator · Caltech Pedestrian (trained on KITTI) | MSE: 0.00307 -> 0.00291 | Improved |
| add flow discriminator (GAN) to flow generator (ablation) | flow generator · Caltech Pedestrian (trained on KITTI) | MSE: 0.00292 -> 0.00289 | Improved |
| predict flows and warp instead of directly generating frames (with GAN) (ablation) | single-generator model · Caltech Pedestrian (trained on KITTI) | MSE: 0.00291 -> 0.00289 | Improved |
| combine frame and flow generators (no GAN) instead of flow GAN alone (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00289 -> 0.00269 | Improved |
| train frame and flow generators jointly, test with flow generator output (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00289 -> 0.00255 | Improved |
| train frame and flow generators jointly, test with frame generator output (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00291 -> 0.00260 | Improved |
| fuse predictions of both generators at test time (ablation) | dual motion GAN · Caltech Pedestrian (trained on KITTI) | MSE: 0.00255 -> 0.00241 | Improved |
| add future-frame prediction during training of flow model (ablation) | dual motion GAN · KITTI flow 2012 (trained on UCF-101) | EPE: 9.3 -> 8.9 | Improved |
| use dual motion GAN pre-training instead of DVF | action recognition · UCF-101 action recognition (split 1) | accuracy: 52.4 -> 55.1 | Improved |
Closest papers by meaning
- Temporal Coherency based Criteria for Predicting Video Frames using Deep Multi-stage Generative Adversarial Networks (NeurIPS 2017 · similarity 0.90)
- Predicting Future Frames Using Retrospective Cycle GAN (CVPR 2019 · similarity 0.89)
- Video Prediction with Appearance and Motion Conditions (ICML 2018 · similarity 0.89)
- Learning to Generate Time-Lapse Videos Using Multi-Stage Dynamic Generative Adversarial Networks (CVPR 2018 · similarity 0.88)
- Visual Dynamics: Probabilistic Future Frame Synthesis via Cross Convolutional Networks (NeurIPS 2016 · similarity 0.88)
- Hierarchical Video Generation From Orthogonal Information: Optical Flow and Texture (AAAI 2018 · 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.
