Research map / Video / Video segmentation and tracking
TF-Blender: Temporal Feature Blender for Video Object Detection
Yiming Cui, Liqi Yan, Zhiwen Cao, Dongfang Liu
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 relation module (ablation) | FGFA · ImageNet VID val (FGFA) | mAP: 77.8 -> 78.5 | Improved |
| add feature adjustment module (ablation) | FGFA · ImageNet VID val (FGFA) | mAP: 77.8 -> 78.1 | Improved |
| add feature blender module (ablation) | FGFA · ImageNet VID val (FGFA) | mAP: 77.8 -> 78.3 | Improved |
| add all three modules (ablation) | FGFA · ImageNet VID val (FGFA) | mAP: 77.8 -> 79.3 | Improved |
| add feature blender to temporal relation (ablation) | FGFA + temporal relation · ImageNet VID val (FGFA) | mAP: 78.5 -> 78.8 | Improved |
| use f_i, f_j, f_i - f_j, f_j - f_i as relation function instead of f_i, f_j (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 78.3 -> 79.3 | Improved |
| use f_i + f_j as relation function instead of f_i - f_j (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 78.6 -> 77.9 | Hurt |
| add f_i + f_j to f_i, f_j relation function (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 78.3 -> 78.1 | Hurt |
| add f_j - f_i to f_i, f_j, f_i - f_j (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 78.9 -> 79.3 | Improved |
| increase mini-network from 1 to 3 conv layers (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 78.8 -> 79.2 | Improved |
| increase mini-network from 3 to 4 conv layers (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 79.2 -> 79.1 | Hurt |
| use 3x3 instead of 1x1 kernels in mini-network (ablation) | FGFA + TF-Blender · ImageNet VID val (FGFA) | mAP: 79.2 -> 79.3 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Stem-Seg — STEm-Seg: Spatio-temporal Embeddings for Instance Segmentation in Videos (ECCV 2020) · 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
- PTSEFormer: Progressive Temporal-Spatial Enhanced TransFormer towards Video Object Detection (ECCV 2022 · similarity 0.87)
- TGBFormer: Transformer-GraphFormer Blender Network for Video Object Detection (AAAI 2025 · similarity 0.86)
- Leveraging Long-Range Temporal Relationships Between Proposals for Video Object Detection (ICCV 2019 · similarity 0.86)
- Temporal Context Enhanced Feature Aggregation for Video Object Detection (AAAI 2020 · similarity 0.85)
- Context Enhanced Transformer for Single Image Object Detection in Video Data (AAAI 2024 · similarity 0.85)
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.84)
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.
