Research map / Video / Video-language understanding
EagleNet: Energy-Aware Fine-Grained Relationship Learning Network for Text-Video Retrieval
Yuhan Chen, Pengwen Dai, Chuan Wang, Dayan Wu, Xiaochun Cao
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 Fine-Grained Relationship Learning (frame-frame relations) (ablation) | base text-frame model · DiDeMo (ViT-B/16) | R@1: 42.1 -> 47.9 | Improved |
| add FRL (ablation) | base text-frame model · MSRVTT (ViT-B/16) | R@1: 48.5 -> 48.8 | Improved |
| add Energy-Aware Matching alone (ablation) | base text-frame model · MSRVTT / DiDeMo | R@1: MSRVTT 48.5 -> 49.0; DiDeMo 42.1 -> 43.4 | Improved |
| add EAM on top of FRL (ablation) | FRL · MSRVTT (ViT-B/16) | R@1: 48.8 -> 50.5 | Improved |
| replace softmax contrastive loss with sigmoid loss in full model (ablation) | EagleNet (FRL+EAM) · DiDeMo (ViT-B/16) | R@1: 49.2 -> 51.5 | Improved |
| replace softmax contrastive loss with sigmoid loss in full model (ablation) | EagleNet (FRL+EAM) · MSRVTT (ViT-B/16) | R@1: 50.5 -> 51.0 | Improved |
| use sigmoid loss alone on base model (ablation) | base text-frame model · MSRVTT (ViT-B/16) | R@1: 48.5 -> 47.8 | Hurt |
| use bilinear energy instead of cosine similarity (ablation) | EAM · MSRVTT (ViT-B/16) | R@1: 50.0 -> 51.0 | Improved |
| use MLP energy instead of bilinear (ablation) | EAM · DiDeMo (ViT-B/16) | R@1: 51.3 -> 51.5 | Mixed |
| average text-frame energies instead of max/min pooling (ablation) | EAM · MSRVTT / DiDeMo | R@1: MSRVTT 50.4 -> 51.0; DiDeMo 51.3 -> 51.5 | Improved |
| compute energy on fused video embedding instead of frames (ablation) | EAM · MSRVTT (ViT-B/16) | R@1: 51.0 -> 48.8 (below w/o EAM 49.9) | Hurt |
| stochastic text modeling without frame context (TMASS) versus XPool | XPool · MSRVTT (ViT-B/16) | R@1: 49.2 -> 48.5 | Hurt |
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.
- XPool — X-Pool: Cross-Modal Language-Video Attention for Text-Video Retrieval (CVPR 2022) · 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
- Towards Efficient and Effective Text-to-Video Retrieval with Coarse-to-Fine Visual Representation Learning (AAAI 2024 · similarity 0.87)
- Rebalancing Contrastive Alignment with Bottlenecked Semantic Increments in Text-Video Retrieval (NeurIPS 2025 · similarity 0.86)
- Text Is MASS: Modeling as Stochastic Embedding for Text-Video Retrieval (CVPR 2024 · similarity 0.86)
- Text Proxy: Decomposing Retrieval from a 1-to-N Relationship into N 1-to-1 Relationships for Text-Video Retrieval (AAAI 2025 · similarity 0.86)
- Stitching Segments and Sentences towards Generalization in Video-Text Pre-training (AAAI 2024 · similarity 0.86)
- Mitigating Semantic Collapse in Partially Relevant Video Retrieval (NeurIPS 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.
