Research map / Video / Video segmentation and tracking
Gradient Forward-Propagation for Large-Scale Temporal Video Modelling
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 Skip-Sideways shortcut connections and temporal integration | Sideways with VGG8 · HMDB51 | accuracy (%): 27.3 -> 31.4 | Helped |
| Add Skip-Sideways shortcut connections and temporal integration | Sideways with VGG8 · UCF101 | accuracy (%): 62.5 -> 64.5 | Helped |
| Add Skip-Sideways shortcut connections and temporal integration | Sideways with VGG16 · Kinetics-600 | accuracy (%): 55.7 -> 61.8 | Helped |
| Use Skip-Sideways instead of causal 3D VGG16 | Causal Kinetics-600 action recognition · Kinetics-600 | accuracy: $64.2 -> $67.3 | Helped |
| Train with Skip-Sideways rather than Sideways | Full-Res future-frame predictor · Kinetics-600 future frame prediction (t+8) | L2-Error: 0.107 -> 0.073 | Helped |
| Use Skip-Sideways model parallelism on 8 devices rather than GPipe on 8 devices | VGG16 training · Kinetics-600 | steps/sec: 4.2 -> \textbf{7.2} | Helped |
| Use concatenation rather than addition for shortcut activation fusion (ablation) | Skip-Sideways · HMDB51 | accuracy (%): 29.6 -> 31.4 | Helped |
| Use concatenation rather than addition for shortcut activation fusion (ablation) | Skip-Sideways · UCF101 | accuracy (%): 62.8 -> 64.5 | Helped |
| Add pseudo-gradient redistribution along shortcut connections (ablation) | FA-Skip-Sideways · HMDB51 | accuracy (%): 29.7 -> 31.4 | Helped |
| Add pseudo-gradient redistribution along shortcut connections (ablation) | FA-Skip-Sideways · UCF101 | accuracy (%): 61.9 -> 64.5 | Helped |
| Add shortcut connections rather than only increasing model width (ablation) | Sideways with VGG8 · HMDB51 | accuracy (%): 27.8 -> 31.4 | Helped |
| Add shortcut connections rather than only increasing model width (ablation) | Sideways with VGG8 · UCF101 | accuracy (%): 64.6 -> 66.0 | Mixed |
Closest papers by meaning
- Stepping Forward on the Last Mile (NeurIPS 2024 · similarity 0.84)
- Frame Context Packing and Drift Prevention in Next-Frame-Prediction Video Diffusion Models (NeurIPS 2025 · similarity 0.84)
- Accelerated training through iterative gradient propagation along the residual path (ICLR 2025 · similarity 0.84)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (ICML 2024 · similarity 0.84)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.84)
- FreeNoise: Tuning-Free Longer Video Diffusion via Noise Rescheduling (ICLR 2024 · similarity 0.83)
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.