Practical CV Wildlife — PCV-to-Wildlife Bridge Skill

Bridge skill connecting the Practical Computer Vision (PCV) course to aerial wildlife detection. 6-agent pipeline that maps PCV curriculum gaps, generates new aerial imagery modules, adapts existing notebooks for wildlife datasets, fills the object detection implementation gap (YOLOv8), creates wildlife-domain exercises, and sequences a combined curriculum. Does NOT re-teach PCV content — generates NEW material that fills gaps and adapts existing exercises for wildlife/aerial imagery. Triggers on: practical computer vision, PCV course, adapt notebook, aerial module, wildlife course, detection module, bridge course, curriculum gap, YOLOv8 exercise, wildlife detection exercise, drone imagery module, adapt PCV, GSD calculation, tile inference exercise, HerdNet module, aerial counting, wildlife curriculum.

Adapt an existing notebook:

Adapt the Pet_Classification.ipynb notebook to use a wildlife camera trap dataset

Fill the detection gap:

Create a hands-on YOLOv8 fine-tuning exercise that builds on PCV Module 8

Full course module:

Build a complete 2-week wildlife detection module

Trigger Conditions

Trigger Keywords

practical computer vision, PCV course, adapt notebook, aerial module, wildlife course, detection module, bridge course, curriculum gap, YOLOv8 exercise, wildlife detection exercise, drone imagery module, adapt PCV, GSD calculation, tile inference exercise, HerdNet module, aerial counting, wildlife curriculum, point-based detection exercise, density estimation exercise, adapt for wildlife, aerial wildlife, PCV wildlife, PCV bridge, drone detection module, wildlife counting exercise

Does NOT Trigger

Source Repository

The PCV course lives at: /Users/christian/PycharmProjects/hnee/practical-computer-vision

PCV Course Inventory

PCV Module Coverage

Critical Gaps This Skill Fills

1. No aerial/drone imagery treatment -- GSD, nadir view, orthomosaics, motion blur, overlap
2. Object detection severely undercovered -- Module 8 is conceptual only, no hands-on YOLO implementation
3. No counting/density estimation -- HerdNet, FIDT maps, point-based detection absent
4. No tile-based inference for large images -- SAHI, overlapping tiles, stitching not covered
5. No wildlife-specific datasets or domain transfer -- All exercises use standard CV datasets (MNIST, ImageNet, pets)

Agent Team (6 Agents)

Operational Modes (6 Modes)

Mode Selection Logic

"Map PCV against wildlife needs"              -> map-curriculum
"Create an aerial imagery fundamentals module" -> generate-aerial-module
"Build a YOLOv8 fine-tuning exercise"         -> fill-detection-gap
"Adapt Pet_Classification for camera traps"   -> adapt-notebook
"Create a tile inference exercise"            -> create-exercise
"Build a full 4-week wildlife detection unit" -> full-course-module

Orchestration Workflow

Full Course Module Mode (complete pipeline)

User: "Build a complete wildlife detection module for PCV students"
     |
=== Phase 1: CURRICULUM MAPPING ===
     |
     +-> [curriculum_mapper_agent]
         - Inventory PCV modules 1-8 coverage
         - Identify gaps against wildlife detection requirements
         - Classify each gap: missing (new content needed) vs adaptable (existing content rewritable)
         - Output: Gap Analysis Report + Prioritized Skill List
     |
=== Phase 2: SEQUENCING ===
     |
     +-> [module_sequencer_agent]
         - Map PCV prerequisites to wildlife skills
         - Slot new content at optimal curriculum points
         - Define week-by-week schedule
         - Output: Curriculum Plan
     |
=== Phase 3: CONTENT GENERATION (parallel where possible) ===
     |
     |-> [aerial_concepts_agent] -> Aerial Imagery Module
     |   - GSD, nadir, overlap, motion blur, footprint
     |   - Notebook: GSD calculation + tile size selection
     |
     |-> [detection_bridge_agent] -> Detection Implementation Module
     |   - YOLOv8 fine-tuning on wildlife data
     |   - mAP evaluation + error analysis
     |   - Counting/density estimation intro
     |
     +-> [exercise_generator_agent] -> Exercise Set
         - 3-5 exercises in PCV style
         - Progressive difficulty
         - Wildlife datasets throughout
     |
=== Phase 4: ASSEMBLY ===
     |
     +-> [module_sequencer_agent]
         - Assemble all materials into final curriculum plan
         - Cross-reference with PCV notebooks
         - Verify prerequisite coverage
         - Output: Complete Course Module Package

Single-Notebook Adaptation Mode

User: "Adapt [notebook_name] for wildlife"
     |
     +-> [wildlife_adapter_agent]
         - Read original notebook structure
         - Identify dataset swap points
         - Preserve pedagogical flow
         - Replace datasets with wildlife equivalents
         - Add wildlife-domain commentary
         - Output: Adapted Notebook Template

Thesis Context (Running Case Study)

All generated content uses the master's thesis by Thomas J. Miesner as a running case study, connecting PCV concepts to real-world aerial wildlife detection:

This case study appears in:

• Aerial concepts module: real GSD calculations from Mavic 2 Pro specs
• Detection bridge: comparison of YOLO vs HerdNet for dense small objects
• Exercise generator: exercises use thesis imagery parameters
• Curriculum plan: thesis as capstone project context

Agent File References

Reference Files

Templates

Integration with Other Skills

practical-cv-wildlife + wildlife-classification -> Classification fine-tuning with wildlife focus
practical-cv-wildlife + herdnet-training        -> From PCV foundations to HerdNet implementation
practical-cv-wildlife + sahi-inference           -> Tile-based inference exercises with real pipeline
practical-cv-wildlife + megadetector             -> MegaDetector as detection baseline exercise
practical-cv-wildlife + active-learning-wildlife -> HITL annotation exercises
practical-cv-wildlife + academic-paper           -> Write course development paper

Quality Standards

Content Quality

1. No PCV duplication -- never re-explain concepts already covered in PCV modules; reference them
2. Prerequisite accuracy -- every new concept must state which PCV module it builds on
3. Wildlife authenticity -- all examples use real species, real sensor specs, real detection challenges
4. Thesis grounding -- case study data comes from the actual thesis, not fabricated numbers

Exercise Quality

5. PCV style fidelity -- exercises follow the same structure as existing PCV notebooks
6. Progressive difficulty -- exercises within a module build on each other
7. Runnable code -- all code cells must be executable (with dataset download instructions)
8. TODO scaffolding -- student tasks are clearly marked with # TODO: comments

Curriculum Quality

9. Prerequisite chain intact -- no wildlife module requires concepts not yet taught in PCV
10. Time estimates realistic -- each module includes estimated completion time
11. Assessment aligned -- exercises test the learning objectives stated at module start

Version History