INDUSTRY FIELD GUIDE
Commercial fishing deck video for AI catch review
One fish can appear in several video frames and then disappear behind equipment.
NAICS 114111 · Finfish Fishing
One fish can appear in several video frames and then disappear behind equipment. A detector might count each appearance as a separate fish. NOAA Fisheries describes electronic monitoring and has published work on automated video processing for fish sampling. A vessel or monitoring provider can discuss separately authorized footage for a model that links detections across views. NOAA Fisheries electronic monitoring · NOAA automated fish video research
Use cleared historical imagery or a staged example with props. Assign consistent track IDs and mark where continuity is known or uncertain. Include one moment when similar specimens overlap and one with too little detail for a species judgment. Test detection, tracking, and uncertainty separately. Never infer a catch total by summing raw detections.
Confirm who controls the footage and monitoring agreement. Keep confidential fishing positions, vessel identifiers, crew images, and regulated recordings out of the first sample. Use practice footage if rights or disclosure authority are unclear.
Describe an approved deck-video dataset
Original Fiund exercise
Reappearance after an occlusion does not automatically add another fish
A staged prop sequence has original frame and track annotations. It represents neither a real catch nor regulated monitoring footage and cannot support a species or vessel assessment.
| Record | Source | What it records |
|---|---|---|
| FD1 | Frames 1–2 | Prop fish F1 visible in frame 1; frame 2 shows F1 partly behind a box. Reviewer confirms continuity. |
| FD2 | Frames 3–4 | F1 reappears with a reviewer-confirmed track link. Another similar prop F2 overlaps it in frame 4; identities remain distinct in the review key. |
| FD3 | Frame 5 | One blurred fish-like shape lacks sufficient features for identity or species. Detector fixture gives one detection but no supported track link. |
Question for the model
How many confirmed unique prop identities appear in frames 1–4, and can frame 5 be added as another fish or assigned a species?
Show the reference answer and checks
FD1 and FD2 establish two unique prop identities, F1 and F2, across frames 1–4. F1's reappearance is a continuation, not another individual. FD3 adds an unresolved detection without identity or species evidence, so it cannot automatically increase the confirmed unique count or receive a species label. Preserve detection, track association and classification as separate outputs rather than deriving a catch total by adding frame detections.
- Counting every frame appearance as a distinct fish.
- Forcing a track or species assignment when the reviewer marks the image insufficient.
Fields to preserve
- Frame time and detection region
- Localizes the visible item.
- Track ID and association confidence
- Supports unique-object counting.
- Occlusion and species-review state
- Records unresolved visual evidence.
Test another case
Introduce two identical props crossing behind an opaque screen with no continuity evidence. Require uncertain post-occlusion identities instead of assumed tracks.
Use staged props or separately cleared historical footage. Exclude vessel and crew identities, positions and regulated recordings; monitoring agreements may control reuse.
Discuss a contribution
Describe the sources you control, the task they could support and what has already been checked. Use the existing-data assessment for an archive or the collection brief for a new recording project. The contributor agreement defines Fiund’s licensing authority and compensation. A guide or fictional exercise does not establish an accepted collection or a sale.