Research map / Video / Video segmentation and tracking
Evolving Space-Time Neural Architectures for Videos
AJ Piergiovanni, Anelia Angelova, Alexander Toshev, Michael S. Ryoo
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 iTGM layers to the search space instead of replacing them with (2+1)D (ablation) | EvaNet · Kinetics-400 | accuracy: 76.6 -> 77.2 | Improved |
| allow iTGM layers in the architecture search (ablation) | EvaNet · Kinetics-400 | accuracy: 76.8 -> 77.2 | Improved |
| lengthen iTGM temporal kernel from L=3 to L=11 without adding parameters (ablation) | iTGM Conv · Kinetics-400 | accuracy: 74.4 -> 74.9 | Improved |
| stretch Kinetics-trained iTGM baseline kernels from L=3 to L=11 (ablation) | iTGM baseline · Charades | mAP: 33.8 -> 34.2 | Improved |
| stretch Kinetics EvaNet iTGM kernels from L=3 to L=11 (ablation) | Kinetics EvaNet · Charades | mAP: 37.7 -> 38.1 | Improved |
| evolve architectures instead of manually designed 3D Conv | video CNN · Kinetics-400 | accuracy: 72.6 -> 76.4 | Improved |
| ensemble three evolved models (EvaNet) instead of best single evolved model | EvaNet · Kinetics-400 | accuracy: 76.4 -> 77.2 | Improved |
| use evolved ensemble instead of a diverse ensemble of baselines | EvaNet · Kinetics-400 | accuracy: 75.3 -> 77.2 | Improved |
| combine Inception-like and ResNet-like meta-architectures | EvaNet · Kinetics-400 | accuracy: 77.2 -> 77.4 | Improved |
| evolve on Kinetics rather than use best non-evolved model, transferred to Moments in Time | EvaNet · Moments in Time | accuracy: 30.7 -> 31.8 | Improved |
| evolve architecture on Charades rather than Kinetics, evaluated on Kinetics | EvaNet · Kinetics-400 | accuracy: 77.2 -> 76.5 | Hurt |
| random ensemble of architectures instead of evolved ensemble | ensemble · Kinetics-400 | accuracy: 77.2 -> 72.6 | Hurt |
Closest papers by meaning
- AssembleNet: Searching for Multi-Stream Neural Connectivity in Video Architectures (ICLR 2020 · similarity 0.88)
- Leaping Into Memories: Space-Time Deep Feature Synthesis (ICCV 2023 · similarity 0.88)
- Evolving Search Space for Neural Architecture Search (ICCV 2021 · similarity 0.87)
- Rethinking Spatiotemporal Feature Learning: Speed-Accuracy Trade-offs in Video Classification (ECCV 2018 · similarity 0.86)
- NEvo: Neural-Guided Evolutionary Video Synthesis for Dynamic Visual Selectivity (NeurIPS 2026 · similarity 0.85)
- einspace: Searching for Neural Architectures from Fundamental Operations (NeurIPS 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.
