Research map / Video / Action recognition
Frame2Freq: Spectral Adapters for Fine-Grained Video Understanding
Thinesh Thiyakesan Ponbagavathi, Constantin Seibold, Alina Roitberg
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 frequency conv. to temporal conv. branch (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 90.4 -> 92.2 | Improved |
| add temporal conv. to frequency-only branch (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 90.9 -> 92.2 | Improved |
| add temporal conv. to frequency-only (ablation) | Frame2Freq-ST · Drive&Act | accuracy: 81.6 -> 82.0 | Improved |
| add frequency conv. to temporal-only (ablation) | Frame2Freq-ST · Drive&Act | accuracy: 75.2 -> 82.0 | Improved |
| place adapter after MHSA instead of before (ablation) | Frame2Freq · IKEA-ASM | accuracy: 76.1 -> 78.1 | Improved |
| place adapter before+after MHSA vs after only (ablation) | Frame2Freq-MS · SSv2 | top-1 accuracy: 68.3 -> 69.7 | Improved |
| mean/concat fusion instead of gated fusion (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 91.7 -> 92.2 | Improved |
| use three window scales instead of one (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 91.5 -> 92.2 | Improved |
| add fourth window scale T/8 (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 92.2 -> 91.0 | Hurt |
| place adapters in all depth ranges vs only layers 1-4 (ablation) | Frame2Freq-MS · Diving48 | top-1 accuracy: 67.6 -> 92.2 | Improved |
| multi-scale variant instead of single-scale ST (ablation) | Frame2Freq · Diving48 | top-1 accuracy: 75.1 -> 92.2 | Improved |
| replace ST-Adapter by Frame2Freq-MS | ST-Adapter · SSv2 | top-1 accuracy: 69.5 -> 70.4 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- ST-Adapter — ST-Adapter: Parameter-Efficient Image-to-Video Transfer Learning (NeurIPS 2022) · better on 5 of 5 shared results
- D2ST-Adapter — D2ST-Adapter: Disentangled-and-Deformable Spatio-Temporal Adapter for Few-shot Action Recognition (ICCV 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- D2ST-Adapter: Disentangled-and-Deformable Spatio-Temporal Adapter for Few-shot Action Recognition (ICCV 2025 · similarity 0.88)
- Where Do We (Not) Need Temporal Context in Low-Resource Video Task Adaptation? (NeurIPS 2026 · similarity 0.86)
- Spectral Motion Alignment for Video Motion Transfer Using Diffusion Models (AAAI 2025 · similarity 0.86)
- MoMa: Modulating Mamba for Adapting Image Foundation Models to Video Recognition (ICML 2025 · similarity 0.85)
- MotionBench: Benchmarking and Improving Fine-grained Video Motion Understanding for Vision Language Models (CVPR 2025 · similarity 0.85)
- Improving LLM Video Understanding with 16 Frames Per Second (ICML 2025 · similarity 0.85)
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.
