Research map / Large language models / AI and society
CFC26: Building Evaluations for Deployment in Sonar-Based Fish Counting
Madison Van Horn, Suzanne Stathatos, Sevan Brodjian, Justin Kay, Kai Van Brunt, Erik Young, Sara Beery, Pietro Perona, Michael Hobley
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Introduces CFC26, an evaluation framework for automated fish counting from sonar videos aimed at field deployment.
Problem
Manual expert inspection of sonar videos to count fish populations is slow, expensive, and hard to reproduce, and existing automated counting systems lack deployment-oriented evaluation.
Approach
Builds evaluations (benchmark framing) for sonar-based fish counting intended to reflect real deployment conditions; the abstract is truncated so specifics are unknown.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add target-river data to multi-river training (Balanced ID vs OOD) (ablation) | YOLOv5m + ByteTrack · Mad | nMAE: 0.800 -> 0.333 | Improved |
| add target-river data to multi-river training (ablation) | YOLOv5m + ByteTrack · Dungeness | nMAE: 0.800 -> 0.000 | Improved |
| add target-river data to multi-river training (ablation) | YOLOv5m + ByteTrack · Eel | nMAE: 1.000 -> 0.800 | No clear effect |
| add multi-river data on top of target-river-only training (ablation) | YOLOv5m + ByteTrack · Chemainus | nMAE: 0.49 -> 0.39 | Improved |
| train on target river only instead of all rivers (ablation) | YOLOv5m + ByteTrack · Nanaimo | nMAE: 0.120 -> 0.627 | Hurt |
| balanced ID training instead of selecting by nMAE: directional bias at Chemainus | YOLOv5m + ByteTrack · Chemainus | nMANE: nMAE 0.39 but nMANE -1.0 | Hurt |
| restrict training to CFC22 locations (geographically close to Dungeness) (ablation) | YOLOv5m + ByteTrack · Dungeness | nMAE: 0.80 -> 0.20 | Improved |
| restrict training to CFC22 locations (ablation) | YOLOv5m + ByteTrack · Eel | nMAE: nMAE approx 1.0 | Hurt |
| use 10k upsampled frames per location instead of no upsampling (ablation) | YOLOv5m · CFC26 sampling comparison | nMAE: 0.11 -> 0.07 | Improved |
| replace YOLOv5m with RT-DETRv2-S (ablation) | detector · CFC26 test | nMAE: 0.0802 -> 0.0906 | No clear effect |
| replace YOLOv5 with YOLO26 (ablation) | detector · CFC26 val | mAP50: 0.6926 -> 0.6610 (medium) | Hurt |
| vary ByteTrack hyperparameters (max age, min hits, IoU threshold) (ablation) | ByteTrack · CFC26 all locations | nMAE: avg relative change 0.93% (min hits), 1.39% (IoU), 4.99% (max age) | No clear effect |
Closest papers by meaning
- Fish-Vista: A Multi-Purpose Dataset for Understanding & Identification of Traits from Images (CVPR 2025 · similarity 0.82)
- DISCount: Counting in Large Image Collections with Detector-Based Importance Sampling (AAAI 2024 · similarity 0.81)
- ReefNet: A Large-Scale Dataset and Benchmark for Fine-Grained Coral Reef Recognition (NeurIPS 2026 · similarity 0.81)
- Combining Observational Data and Language for Species Range Estimation (NeurIPS 2024 · similarity 0.80)
- DivShift: Exploring Domain-Specific Distribution Shifts in Large-Scale, Volunteer-Collected Biodiversity Datasets (AAAI 2025 · similarity 0.80)
- BirdCollect: A Comprehensive Benchmark for Analyzing Dense Bird Flock Attributes (AAAI 2024 · similarity 0.80)
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.
