Research map / Video / Video-language understanding
The "Something Something" Video Database for Learning and Evaluating Visual Common Sense
Raghav Goyal, Samira Ebrahimi Kahou, Vincent Michalski, Joanna Materzynska, Susanne Westphal, Heuna Kim, Valentin Haenel, Ingo Fruend, Peter Yianilos, Moritz Mueller-Freitag, Florian Hoppe, Christian Thurau and 2 more
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 |
|---|---|---|---|
| initialize 2D CNN with ImageNet pretraining | VGG-16 + Avg · Something-Something (10 classes) | top-1 error (%): 76.5 -> 54.7 | Improved |
| initialize 2D CNN with ImageNet pretraining | VGG-16 + Avg · Something-Something (40 classes) | top-1 error (%): 88.0 -> 79.2 | Improved |
| replace frame averaging with LSTM | Pre-2D CNN · Something-Something (10 classes) | top-1 error (%): 54.7 -> 52.3 | Improved |
| replace frame averaging with LSTM | Pre-2D CNN · Something-Something (40 classes) | top-1 error (%): 79.2 -> 77.8 | Improved |
| use 3D CNN instead of 2D CNN, both from scratch | scratch baselines · Something-Something (40 classes) | top-1 error (%): 88.0 -> 70.3 | Improved |
| use Sports-1M pretrained 3D CNN instead of pretrained 2D CNN | pretrained baselines · Something-Something (10 classes) | top-1 error (%): 54.7 -> 47.5 | Improved |
| combine 2D and 3D CNN encodings | best 3D CNN · Something-Something (10 classes) | top-1 error (%): 47.5 -> 44.9 | Improved |
| combine 2D and 3D CNN encodings | best 3D CNN · Something-Something (40 classes) | top-1 error (%): 66.2 -> 63.8 | Improved |
| increase number of classes from 10 to 40 to 174 | Pre-3D CNN + Avg · Something-Something | top-1 error (%): 47.5 -> 66.2 -> 88.5 | Hurt |
| add dropout on first FC layer and batch norm on last layer (ablation) | baseline models · Something-Something | training | Improved |
| add flipping and small random rotation augmentation (ablation) | baseline models · Something-Something | accuracy | No clear effect |
Closest papers by meaning
- Looking Beyond the Obvious: A Survey on Abstract Concept Recognition for Video Understanding (IJCV 2026 · similarity 0.88)
- PerceptionLM: Open-Access Data and Models for Detailed Visual Understanding (NeurIPS 2025 · similarity 0.86)
- Perceive Anything: Recognize, Explain, Caption, and Segment Anything in Images and Videos (NeurIPS 2025 · similarity 0.86)
- Look, Remember and Reason: Grounded Reasoning in Videos with Language Models (ICLR 2024 · similarity 0.86)
- Simple Unsupervised Object-Centric Learning for Complex and Naturalistic Videos (NeurIPS 2022 · similarity 0.86)
- Black Swan: Abductive and Defeasible Video Reasoning in Unpredictable Events (CVPR 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.
