Research map / Video / Action recognition
LSTA: Long Short-Term Attention for Egocentric 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 output pooling to the ConvLSTM baseline (ablation) | ConvLSTM · GTEA 61 fixed split | accuracy (%): 51.72 -> 62.07 | Improved |
| Add attention pooling to the ConvLSTM baseline (ablation) | ConvLSTM · GTEA 61 fixed split | accuracy (%): 51.72 -> 66.38 | Improved |
| Add both attention pooling and output pooling (ablation) | ConvLSTM · GTEA 61 fixed split | accuracy (%): 51.72 -> 68.1 | Improved |
| Add bias control to attention and output pooling (ablation) | LSTA · GTEA 61 fixed split | accuracy (%): 68.1 -> 74.14 | Improved |
| Use cross-modal fusion instead of late fusion (ablation) | LSTA two-stream network · GTEA 61 fixed split | accuracy (%): 78.45 -> 79.31 | Improved |
| Use LSTA rather than ego-rnn | GTEA 61 fixed split · GTEA 61 fixed split | accuracy (%): 63.79 -> 74.14 | Improved |
| Use attention on the motion stream (ablation) | Action-pretrained ResNet-34 flow network · GTEA 61 fixed split | accuracy (%): 36.21 -> 40.52 | Improved |
| Use the LSTA two-stream model rather than the strongest reported TSN baseline | EPIC-KITCHENS · EPIC-KITCHENS S1 | Top-1 action accuracy (%): 20.54 -> 30.33 | Improved |
| Use the LSTA two-stream model rather than the strongest reported TSN baseline | EPIC-KITCHENS · EPIC-KITCHENS S2 | Top-1 action accuracy (%): 10.89 -> 16.63 | Improved |
| Use recurrent attention tracking rather than frame-independent attention | LSTA compared with ego-rnn · GTEA 61 fixed split | accuracy (%): 63.79 -> 74.14 | Improved |
Later papers that compared with this method
3 later accepted papers report LSTA as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Learning to Recognize Actions on Objects in Egocentric Video With Attention Dictionaries (TPAMI 2021 · better on 2 of 2 shared results)
- Symbiotic Attention with Privileged Information for Egocentric Action Recognition (AAAI 2020 · better on 1 of 1 shared result)
- What Would You Expect? Anticipating Egocentric Actions With Rolling-Unrolling LSTMs and Modality Attention (ICCV 2019 · better on 1 of 1 shared result)
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
- Egocentric Action Recognition by Automatic Relation Modeling (TPAMI 2022 · similarity 0.89)
- Rolling-Unrolling LSTMs for Action Anticipation from First-Person Video (TPAMI 2020 · similarity 0.89)
- Learning to Recognize Actions on Objects in Egocentric Video With Attention Dictionaries (TPAMI 2021 · similarity 0.89)
- Action Scene Graphs for Long-Form Understanding of Egocentric Videos (CVPR 2024 · similarity 0.89)
- Interactive Prototype Learning for Egocentric Action Recognition (ICCV 2021 · similarity 0.88)
- Ego-Only: Egocentric Action Detection without Exocentric Transferring (ICCV 2023 · 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.
