Research map / Video / Action recognition
Gate-Shift Networks for Video Action Recognition
Swathikiran Sudhakaran, Sergio Escalera, Oswald Lanz
Check 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 GSM to TSN (ablation) | BN-Inception TSN · Something-V1 validation | accuracy (%): 17.25 -> 47.24 | Improved |
| Increase GSM count from 1 to 5 to 10 (ablation) | BN-Inception · Something-V1 validation | accuracy (%): 22.7 -> 43.13 -> 47.24 | Improved |
| Place GSM in branch 4 rather than another single branch (ablation) | BN-Inception · Something-V1 validation | accuracy (%): Branch 1: 45.11; Branch 2: 44.98; Branch 3: 45.05; Branch 4: 47.24 | Improved |
| Place GSM in all branches instead of the selected branch (ablation) | BN-Inception · Something-V1 validation | accuracy (%): 47.24 -> 43.5 | Hurt |
| Replace tanh gate activation with sigmoid (ablation) | GSM · Something-V1 validation | accuracy (%): 47.24% vs 44.75% | Hurt |
| Reverse the order of input video frames (ablation) | GSM BN-Inception · Something-V1 validation | accuracy (%): 47.24 to 15.38% | Hurt |
| Use InceptionV3 instead of BN-Inception | GSM · Something-V1 | accuracy (%): 49.56 -> 50.63 | Improved |
| Ensemble models trained with different frame counts (ablation) | GSM InceptionV3 · Something-V1 | accuracy (%): 52.57 -> 54.04 -> 54.88 -> 55.16 | Improved |
| Increase input frame count from 8 to 12 (ablation) | GSM InceptionV3 · Something-V1 | accuracy (%): 49.01 -> 51.58 | Improved |
| Increase input frame count from 12 to 16 or 24 (ablation) | GSM InceptionV3 · Something-V1 | accuracy (%): 51.58 -> 50.63; 51.58 -> 49.63 | Mixed |
| Use GSM rather than the strongest listed Diving48 baseline | GSM · Diving48 | accuracy (%): 38.8 -> 40.27 | Improved |
Closest papers by meaning
- Gate-Shift-Fuse for Video Action Recognition (TPAMI 2023 · similarity 0.93)
- Deep Analysis of CNN-Based Spatio-Temporal Representations for Action Recognition (CVPR 2021 · similarity 0.89)
- A Closer Look at Spatiotemporal Convolutions for Action Recognition (CVPR 2018 · similarity 0.87)
- Learning SpatioTemporal and Motion Features in a Unified 2D Network for Action Recognition (TPAMI 2022 · similarity 0.87)
- TSM: Temporal Shift Module for Efficient and Scalable Video Understanding on Edge Devices (TPAMI 2020 · similarity 0.87)
- ACTION-Net: Multipath Excitation for Action Recognition (CVPR 2021 · 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.
