Research map / Video / Action recognition
Deep Alternative Neural Network: Exploring Contexts as Early as Possible for Action Recognition
Read the paperCheck 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 recurrent layers (alternative layers) stacked with volumetric conv, 6 ALs vs 6 VC layers (ablation) | DANN · UCF101 split1 | video-level accuracy: 80.2 -> 87.9 | Helped |
| increase number of recurrent unfolding steps T from 3 to 6 (ablation) | DANN 6AL_VPP_3FC · UCF101 split1 | video-level accuracy: 87.9 -> 89.0 | Helped |
| place alternative layers early vs late in network (ablation) | DANN · UCF101 split1 | video-level accuracy: AL-last 80.9 vs AL-early 87.9 | Helped |
| adaptive optical-flow-based temporal input length with VPPL instead of fixed 16-frame clips (ablation) | DANN · UCF101 split1 | video-level accuracy: 77.6 -> 83.0 | Helped |
| use TVL1 optical flow instead of RGB as input (ablation) | DANN input modality · UCF101 split1 | video-level accuracy (clip-level for TVL1 78.1): RGB 64.9 -> TVL1 79.6 | Helped |
| combined random clipping and multi-scale clipping augmentation (ablation) | DANN · UCF101 split1 | video-level accuracy: multi-scale alone 82.4 vs combined 82.3 video-level; combined clip-level 81.6 vs 81.2 | Mixed |
| add spatial RGB stream fusion (0.4/0.6 weighting) | DANN · HMDB51 and UCF101 (3 splits) | video-level accuracy: HMDB51 63.3 -> 65.9; UCF101 89.2 -> 91.6 | Helped |
| fine-tune from another dataset (UCF101 to HMDB51) instead of scratch (ablation) | DANN · HMDB51 split1 | accuracy: 56.4 -> 62.5 | Helped |
Closest papers by meaning
- Deep Sequential Context Networks for Action Prediction (CVPR 2017 · similarity 0.88)
- VideoNet: A Large-Scale Dataset for Domain-Specific Action Recognition (CVPR 2026 · similarity 0.88)
- Dynamic Normalization and Relay for Video Action Recognition (NeurIPS 2021 · similarity 0.87)
- More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation (NeurIPS 2019 · similarity 0.87)
- Prompt-guided Disentangled Representation for Action Recognition (NeurIPS 2025 · similarity 0.87)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · 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.