Research map / Video / Video segmentation and tracking
Context-Driven Detection of Invertebrate Species in Deep-Sea Video
R. Austin McEver, Bowen Zhang, Connor Levenson, Adnan Iftekhar, Bangalore S Manjunath
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 Negative Region Dropping (rho 0.75) (ablation) | Faster RCNN (lr 0.01) · DUSIA test (annotated frames) | mAP@0.5: 0.391 -> 0.439 | Improved |
| set Negative Region Dropping rho to 1 (ablation) | Faster RCNN (lr 0.01) · DUSIA test (annotated frames) | mAP@0.5: 0.391 -> 0.264 | Hurt |
| add global feature fusion (beta 0.0001) (ablation) | Faster RCNN (lr 0.01) · DUSIA test (annotated frames) | mAP@0.5: 0.391 -> 0.410 | Improved |
| use too large fusion scale beta 0.1 (ablation) | Faster RCNN (lr 0.01) · DUSIA test (annotated frames) | mAP@0.5: 0.391 -> 0.371 | Hurt |
| add explicit context classification (alpha 0.0001) (ablation) | Faster RCNN (lr 0.01) · DUSIA test (annotated frames) | mAP@0.5: 0.391 -> 0.420 | Improved |
| add Negative Region Dropping to the explicit context detector (ablation) | CDD alpha 0.0001 · DUSIA test (annotated frames) | mAP@0.5: 0.420 -> 0.439 | Improved |
| combine implicit and explicit context with NRD (ablation) | CDD · DUSIA test (annotated frames) | mAP@0.5: 0.430 | No clear effect |
| add context to detector (global fusion) (ablation) | Faster RCNN · DUSIA test (annotated frames) | mAP@0.5 relative: 7.4% increase | Improved |
| add Negative Region Dropping (ablation) | Faster RCNN · DUSIA test (annotated frames) | mAP@0.5 relative: 12.3% | Improved |
| add both context and Negative Region Dropping (ablation) | Faster RCNN · DUSIA test (annotated frames) | mAP@0.5 relative: 14.3% increase | Improved |
| lower learning rate to 0.001 with NRD rho 0.75 (ablation) | Faster RCNN · DUSIA test (annotated frames) | mAP@0.5: 0.439 -> 0.380 | Hurt |
| train separate binary classifier per substrate instead of one multi-class classifier | ResNet-50 substrate classifier · DUSIA substrate classification (test_wv) | mAP: 0.663 -> 0.692 | Improved |
Closest papers by meaning
- ReefNet: A Large-Scale Dataset and Benchmark for Fine-Grained Coral Reef Recognition (NeurIPS 2026 · similarity 0.87)
- Fantastic Animals and Where to Find Them: Segment Any Marine Animal with Dual SAM (CVPR 2024 · similarity 0.86)
- SeafloorAI: A Large-scale Vision-Language Dataset for Seafloor Geological Survey (NeurIPS 2024 · similarity 0.85)
- Enhancing Vision-Language Models with Morphological and Taxonomic Knowledge: Towards Coral Recognition for Ocean Health (AAAI 2025 · similarity 0.85)
- Fish-Vista: A Multi-Purpose Dataset for Understanding & Identification of Traits from Images (CVPR 2025 · similarity 0.84)
- NAUTILUS: A Large Multimodal Model for Underwater Scene Understanding (NeurIPS 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.
