Research map / Video / Video segmentation and tracking
Sideways: Depth-Parallel Training of Video Models
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 |
|---|---|---|---|
| Train VGG-8 with Sideways instead of BP | VGG-8 · HMDB51, first split | accuracy (%): 24.6 -> 25.8 | Helped |
| Train Simple-CNN with Sideways instead of BP | Simple-CNN · HMDB51, first split | accuracy (%): 17.2 -> 16.5 | Hurt |
| Train VGG-8 with Sideways instead of BP | VGG-8 · UCF101 | accuracy (%): 49.1 -> 53.8 | Helped |
| Train Simple-CNN with Sideways instead of BP | Simple-CNN · UCF101 | accuracy (%): 40.7 -> 42.16 | Helped |
| Train VGG-8 with Sideways instead of BP | VGG-8 + Dropout (0.9) · UCF101 | accuracy (%): 56.0 -> 58.2 | Helped |
| Add dropout (0.9) (ablation) | VGG-8 with Sideways · UCF101 | accuracy (%): 53.8 -> 58.2 | Helped |
| Use depth-pipelined Sideways training instead of BP | Simple-CNN · Depth-parallel training | training steps per second: 1.7 -> 8.4; 4.9x | Helped |
| Use depth-pipelined Sideways training instead of BP | VGG-8 · Depth-parallel training | training steps per second: 0.1 -> 0.6; 6.0x | Helped |
| Train the encoder with Sideways instead of BP | Encoder-decoder · CATER all_actions, auto-encoding | mean squared error: 0.014 -> 0.002 | Helped |
| Increase temporal striding to skip frames (ablation) | Sideways and BP video classifiers · UCF101 | accuracy | Mixed |
| Use a relatively high learning rate with Sideways (ablation) | Video classifiers · UCF101 | training accuracy | Hurt |
| Apply a mask to unavailable pipeline gradients (ablation) | Sideways · Video model training | training stability | Helped |
Closest papers by meaning
- Gradient Forward-Propagation for Large-Scale Temporal Video Modelling (CVPR 2021 · similarity 0.89)
- Accelerated training through iterative gradient propagation along the residual path (ICLR 2025 · similarity 0.85)
- PETRA: Parallel End-to-end Training with Reversible Architectures (ICLR 2025 · similarity 0.85)
- Learning Temporally Consistent Video Depth from Video Diffusion Priors (CVPR 2025 · similarity 0.84)
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.84)
- Reducing Fine-Tuning Memory Overhead by Approximate and Memory-Sharing Backpropagation (ICML 2024 · similarity 0.84)
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.