Research map / Video / Action recognition
RubiksNet: Learnable 3D-Shift for Efficient Video Action Recognition
Linxi Fan, Shyamal Buch, Guanzhi Wang, Ryan Cao, Yuke Zhu, Juan Carlos Niebles, Li Fei-Fei
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 |
|---|---|---|---|
| learn spatial shifts instead of fixed [29] (ablation) | RubiksNet-Large · HMDB-51 | top-1 accuracy: 70.0 -> 71.4 | Improved |
| learn temporal shifts instead of fixed [16] (ablation) | RubiksNet-Large · HMDB-51 | top-1 accuracy: 69.5 -> 71.4 | Improved |
| learn both spatial and temporal shifts instead of both fixed (ablation) | RubiksNet-Large · HMDB-51 | top-1 accuracy: 68.3 -> 71.4 | Improved |
| learn both spatial and temporal shifts instead of both fixed (ablation) | RubiksNet-Small · HMDB-51 | top-1 accuracy: 65.7 -> 69.5 | Improved |
| learn temporal shifts instead of fixed [16] (ablation) | RubiksNet-Small · HMDB-51 | top-1 accuracy: 67.8 -> 69.5 | Improved |
| learn spatial shifts instead of fixed [29] (ablation) | RubiksNet-Small · HMDB-51 | top-1 accuracy: 67.5 -> 69.5 | Improved |
| use interpolated quantized (RS-IQ) instead of interpolated shift (ablation) | RubiksShift · Val Top-1 (shift type ablation) | top-1 accuracy: 61.7 -> 58.2 | Hurt |
| use attention quantized (RS-AQ) with 0.125 budget instead of full interpolated (ablation) | RubiksShift · Val Top-1 (shift type ablation) | top-1 accuracy: 61.7 -> 61.6 | No clear effect |
| use attention quantized (RS-AQ) instead of interpolated quantized (RS-IQ) (ablation) | RubiksShift · Val Top-1 (shift type ablation) | top-1 accuracy: 58.2 -> 61.6 | Improved |
| use RubiksNet-Tiny instead of TSM-Small | video recognition · Something-Something-v2 (1-clip) | top-1 accuracy: 49.3 -> 54.6 | Improved |
| use RubiksNet-Large instead of TSM-Large | video recognition · Something-Something-v2 (1-clip) | top-1 accuracy: 58.8 -> 59.0 | No clear effect |
| use RubiksNet instead of TSM | video recognition · HMDB-51 (2-clip, Large) | top-1 accuracy: 73.5 -> 74.6 | Improved |
Closest papers by meaning
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.90)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.89)
- Spatiotemporal Pyramid Network for Video Action Recognition (CVPR 2017 · similarity 0.89)
- T-C3D: Temporal Convolutional 3D Network for Real-Time Action Recognition (AAAI 2018 · similarity 0.89)
- A Large-scale Study of Spatiotemporal Representation Learning with a New Benchmark on Action Recognition (ICCV 2023 · similarity 0.89)
- Sympathy for the Details: Dense Trajectories and Hybrid Classification Architectures for Action Recognition (ECCV 2016 · similarity 0.88)
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.
