Research map / Video / Video segmentation and tracking
SDC-Net: Video prediction using spatially-displaced convolution
Fitsum A. Reda, Guilin Liu, Kevin J. Shih, Robert Kirby, Jon Barker, David Tarjan, Andrew Tao, Bryan Catanzaro
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 |
|---|---|---|---|
| use SDC (displaced kernel) instead of vector-based resampling (ablation) | SDC-Net · Caltech Pedestrian test | SSIM: 0.902 -> 0.918 | Improved |
| use SDC instead of vector-based resampling (ablation) | SDC-Net · Caltech Pedestrian test | L2 (1e-3): 2.47 -> 1.62 | Improved |
| use SDC instead of kernel-based resampling (ablation) | SDC-Net · Caltech Pedestrian test | L2 (1e-3): 2.19 -> 1.62 | Improved |
| use SDC instead of kernel-based resampling (ablation) | SDC-Net · Caltech Pedestrian test | SSIM: 0.896 -> 0.918 | Improved |
| use SDC instead of vector-based resampling (ablation) | SDC-Net · YouTube-8M (26 HD videos) | SSIM: 0.905 -> 0.911 | Improved |
| use SDC instead of vector-based resampling (ablation) | SDC-Net · YouTube-8M (26 HD videos) | L2: 0.00270 -> 0.00240 | Improved |
| use SDC instead of kernel-based resampling (ablation) | SDC-Net · YouTube-8M (26 HD videos) | SSIM: 0.904 -> 0.911 | Improved |
| use SDC instead of kernel-based resampling (ablation) | SDC-Net · YouTube-8M (26 HD videos) | PSNR: 37.33 -> 37.15 | Hurt |
| use SDC instead of vector-based resampling (ablation) | SDC-Net · YouTube-8M (26 HD videos) | PSNR: 37.24 -> 37.15 | Hurt |
| use vector-based instead of kernel-based resampling (ablation) | SDC-Net variants · Caltech Pedestrian test | SSIM: 0.896 -> 0.902 | Improved |
| fine-tune with perceptual and style loss after L1 (ablation) | SDC-Net · qualitative frames | sharpness | Improved |
| train with L1 or L2 loss against estimated optical flow for motion vectors (ablation) | SDC-Net · qualitative | accuracy | Hurt |
Closest papers by meaning
- Flexible Spatio-Temporal Networks for Video Prediction (CVPR 2017 · similarity 0.88)
- Flow-Grounded Spatial-Temporal Video Prediction from Still Images (ECCV 2018 · similarity 0.88)
- Motion Graph Unleashed: A Novel Approach to Video Prediction (NeurIPS 2024 · similarity 0.87)
- TTVidT: Decoupling the Temporal Axis for Efficient Motion-Centric Video Pretraining (NeurIPS 2026 · similarity 0.86)
- Recurrent Partial Kernel Network for Efficient Optical Flow Estimation (AAAI 2024 · similarity 0.86)
- GenRec: Unifying Video Generation and Recognition with Diffusion Models (NeurIPS 2024 · 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.
