Research map / Video / Action recognition
Segmental Spatiotemporal CNNs for Fine-Grained Action Segmentation
Colin Lea, Austin Reiter, René Vidal, Gregory D. Hager
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 temporal 1D convolution over 10 s to the spatial CNN (ablation) | S-CNN · 50 Salads (eval granularity) | edit score: 24.10 -> 60.98 | Improved |
| add temporal 1D convolution over 10 s to the spatial CNN (ablation) | S-CNN · 50 Salads (eval granularity) | frame-wise accuracy: 66.64 -> 71.37 | Improved |
| add temporal 1D convolution to the spatial CNN (ablation) | S-CNN · JIGSAWS | frame-wise accuracy: 67.25 -> 71.21 | Improved |
| add segmental semi-Markov inference on top of ST-CNN (ablation) | ST-CNN · JIGSAWS | edit score: 59.89 -> 66.56 | Improved |
| add segmental semi-Markov inference on top of ST-CNN (ablation) | ST-CNN · 50 Salads (eval granularity) | edit score: 60.98 -> 62.06 | Improved |
| add segmental semi-Markov inference on top of ST-CNN (ablation) | ST-CNN · JIGSAWS | frame-wise accuracy: 71.21 -> 74.22 | Improved |
| use temporal CNN filters instead of LSTM on top of spatial features | S-CNN + LSTM · JIGSAWS | edit score: 54.07 -> 59.89 | Improved |
| use temporal CNN filters instead of LSTM on top of spatial features | S-CNN + LSTM · 50 Salads (eval granularity) | frame-wise accuracy: 66.30 -> 71.37 | Improved |
| add LSTM on top of spatial CNN (ablation) | S-CNN · 50 Salads (eval granularity) | frame-wise accuracy: 66.64 -> 66.30 | No clear effect |
| replace IDT features with the spatial CNN | IDT · 50 Salads (eval granularity) | frame-wise accuracy: 54.28 -> 66.64 | Improved |
| replace pre-trained VGG features with the task-trained spatial CNN | VGG · 50 Salads (eval granularity) | frame-wise accuracy: 38.30 -> 66.64 | Improved |
| use coarser action granularity (high, 4 classes instead of low, 52 classes) | ST-CNN + Seg · 50 Salads | frame-wise accuracy: 44.13 -> 92.43 | Improved |
Closest papers by meaning
- Learning Motion in Feature Space: Locally-Consistent Deformable Convolution Networks for Fine-Grained Action Detection (ICCV 2019 · similarity 0.91)
- Temporal Convolutional Networks for Action Segmentation and Detection (CVPR 2017 · similarity 0.90)
- 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)
- Spatiotemporal Fusion in 3D CNNs: A Probabilistic View (CVPR 2020 · 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.
