Research map / Probabilistic and scientific modeling / Variational and generative sampling
An Uncertain Future: Forecasting from Static Images Using Variational Autoencoders
Jacob Walker, Carl Doersch, Abhinav Gupta, Martial Hebert
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 a latent-variable VAE to the direct regressor | trajectory predictor · THUMOS 2015 trajectory prediction | negative log likelihood: 11463 -> 11082 | Improved |
| use VAE trajectory model instead of optical flow extrapolation | trajectory predictor · THUMOS 2015 trajectory prediction | negative log likelihood: 11734 -> 11082 | Improved |
| sample multiple predictions from the VAE instead of regressor output | trajectory predictor · THUMOS 2015 minimum Euclidean distance | average minimum Euclidean distance | Mixed |
| pretrain on trajectory prediction with calibration instead of training from scratch | Fast-RCNN AlexNet-style · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 40.6 -> 47.3 | Improved |
| pretrain on trajectory prediction instead of kmeans initialization | Fast-RCNN · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 46.1 -> 47.3 | Improved |
| use trajectory prediction pretraining instead of relative position on ImageNet | Fast-RCNN · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 52.3 -> 47.3 | Hurt |
| pretrain on trajectory prediction instead of egomotion | Fast-RCNN · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 44.9 -> 47.3 | Improved |
| add between-layer scale calibration (ablation) | video embedding pretraining (8M frames) · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 47.5 -> 42.2 | Hurt |
| increase video embedding pretraining from 1.5M to 8M frames | video embedding baseline · PASCAL VOC 2012 test detection (Fast-RCNN) | mAP: 46.2 -> 47.5 | Improved |
| use trajectory pretraining instead of relative position | Fast-RCNN · PASCAL VOC 2012 test detection, person class (Fast-RCNN) | AP: 57.5 -> 58.4 | Improved |
| pretrain on trajectory prediction instead of relative position | Fast-RCNN · PASCAL VOC 2012 test detection, chair class (Fast-RCNN) | AP: 27.0 -> 21.2 | Hurt |
Closest papers by meaning
- Probabilistic Future Prediction for Video Scene Understanding (ECCV 2020 · similarity 0.92)
- Compositional Video Prediction (ICCV 2019 · similarity 0.90)
- Schroedinger’s Cat: Probabilistic Representation and Prediction of Potential Scene Kinematics (ECCV 2026 · similarity 0.90)
- DESIRE: Distant Future Prediction in Dynamic Scenes With Interacting Agents (CVPR 2017 · similarity 0.89)
- Advancing Semantic Future Prediction through Multimodal Visual Sequence Transformers (CVPR 2025 · similarity 0.87)
- Towards Understanding Future: Consistency Guided Probabilistic Modeling for Action Anticipation (AAAI 2024 · 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.
