Research map / Video / Video segmentation and tracking
TASED-Net: Temporally-Aggregating Spatial Encoder-Decoder Network for Video Saliency Detection
Kyle Min, Jason J. Corso
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 |
|---|---|---|---|
| use late two-step instead of early two-step aggregation (ablation) | TASED-Net (T=32) · DHF1K val | NSS: 2.673 -> 2.706 | Improved |
| use late two-step instead of late aggregation (ablation) | TASED-Net (T=32) · DHF1K val | NSS: 2.618 -> 2.706 | Improved |
| add Kinetics pre-training of the encoder (ablation) | TASED-Net · DHF1K val | NSS: about 0.5 NSS | Improved |
| add auxiliary pooling with max-unpooling instead of trilinear upsampling (ablation) | TASED-Net · DHF1K val | NSS: 2.452 -> 2.706 | Improved |
| add auxiliary pooling with max-unpooling instead of transposed convolution (ablation) | TASED-Net · DHF1K val | NSS: 2.598 -> 2.706 | Improved |
| increase clip length from 16 to 32 frames (ablation) | TASED-Net · DHF1K val | NSS: 2.622 -> 2.706 | Improved |
| increase clip length from 32 to 64 frames (ablation) | TASED-Net · DHF1K val | NSS: 2.706 -> 2.554 | Hurt |
| reduce clip length from 32 to 4 frames (ablation) | TASED-Net · DHF1K val | NSS: 2.706 -> 2.434 | Hurt |
| add two more stride-1 transposed convolutions per spatial decoding block (denser decoder) (ablation) | TASED-Net · DHF1K val | NSS | Hurt |
| predict multiple saliency maps per window instead of a single map (ablation) | TASED-Net · DHF1K val | NSS | Hurt |
| use TASED-Net over ACLNet | TASED-Net · DHF1K test | NSS: 2.354 -> 2.667 | Improved |
| use TASED-Net over ACLNet | TASED-Net · Hollywood2 test | NSS: 3.086 -> 3.302 | Improved |
Later papers that compared with this method
1 later accepted paper reports TASED-Net as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Measuring the Importance of Temporal Features in Video Saliency (ECCV 2020 · better on 3 of 5 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
- Temporal Saliency Query Network for Efficient Video Recognition (ECCV 2022 · similarity 0.88)
- Salient Object Detection with Recurrent Fully Convolutional Networks (TPAMI 2018 · similarity 0.87)
- Revisiting Video Saliency: A Large-Scale Benchmark and a New Model (CVPR 2018 · similarity 0.87)
- Revisiting Video Saliency Prediction in the Deep Learning Era (TPAMI 2019 · similarity 0.86)
- Adaptive Graph Convolutional Network With Attention Graph Clustering for Co-Saliency Detection (CVPR 2020 · similarity 0.86)
- NSNet: Non-Saliency Suppression Sampler for Efficient Video Recognition (ECCV 2022 · similarity 0.86)
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.
