Project Overview I’m looking for an experienced computer vision / machine learning consultant to help me build a custom object detection dataset and train an Ultralytics YOLO model capable of identifying 22 North American big game animal species. The trained model must be exported to CoreML for deployment in an iOS application. Target Species (22 Classes) 1. White-tailed Deer 2. Mule Deer 3. Coues Deer 4. Sitka Black-tailed Deer 5. Elk (Wapiti) 6. Moose 7. Caribou 8. Pronghorn 9. American Bison 10. Muskox 11. Rocky Mountain Bighorn Sheep 12. Desert Bighorn Sheep 13. Dall Sheep 14. Stone Sheep 15. Mountain Goat 16. American Black Bear 17. Grizzly Bear / Brown Bear 18. Mountain Lion / Cougar 19. Gray Wolf 20. Coyote 21. Bobcat 22. Feral Hog / Wild Boar Scope of Work Phase 1 — Dataset Creation & Annotation • Source and curate images for all 22 species (minimum 500–1,000+ annotated images per class recommended) • Images should represent realistic field conditions: trail cameras, phone cameras, hunting optics, variable lighting, partial occlusion by vegetation, varying distances, and diverse terrain (forest, desert, tundra, mountain, swamp) • Include seasonal variation (summer coat vs. winter coat) and sex/age variation (bulls, cows, calves, bucks, does, fawns, boars, sows, etc.) where visually distinct • Annotate all images with bounding boxes in YOLO format using Roboflow • Apply appropriate preprocessing and augmentation (rotation, brightness, blur, mosaic, etc.) • Merge any relevant existing public datasets from Roboflow Universe where applicable • Generate versioned dataset splits (train/valid/test) Phase 2 — Model Training • Train using Ultralytics YOLO (YOLOv11 or newer recommended architecture) • Optimize for mobile inference (YOLO Nano or Small variant preferred for iOS performance) • Achieve acceptable mAP across all 22 classes; provide confusion matrix, precision-recall curves, and per-class performance metrics • Iterate on problem classes (e.g., distinguishing mule deer from white-tailed deer, or Rocky Mountain bighorn from desert bighorn) Phase 3 — CoreML Export & iOS Validation • Export the trained model to CoreML format using model.export(format=‘coreml’) • Validate inference on sample images to confirm the model runs correctly on-device • Deliver final model weights (.pt), CoreML model (.mlmodel/.mlpackage), dataset (via Roboflow), and training notebook/scripts Deliverables • Annotated dataset hosted on Roboflow with version history • Trained YOLO model weights (best.pt and last.pt) • CoreML exported model ready for iOS integration • Training notebook or scripts (reproducible) • Performance report: mAP, per-class precision/recall, confusion matrix • Documentation of dataset sources, augmentation settings, and training hyperparameters Required Skills & Experience • Proven experience with Roboflow (dataset creation, annotation, augmentation) • Strong hands-on experience training Ultralytics YOLO models on custom datasets • Experience exporting models to CoreML for iOS deployment • Familiarity with wildlife or animal detection datasets is a strong plus • Understanding of challenges specific to wildlife identification (camouflage, occlusion, similar species differentiation, variable image quality) Preferred Qualifications • Portfolio or past project involving animal/wildlife detection • Experience with dataset sourcing strategies (trail cam imagery, open datasets, web scraping pipelines) • Familiarity with iOS/Swift/CoreML integration (not required but a plus) • Knowledge of North American big game species and visual distinguishing features Budget & Timeline • Budget: [Open to proposals — please provide an estimate based on scope] • Timeline: [Flexible — propose a realistic timeline with milestones for each phase] • Engagement Type: Fixed-price with milestone payments preferred (Phase 1, Phase 2, Phase 3) To Apply, Please Include: 1. Relevant portfolio examples (especially any animal/wildlife detection work) 2. Your proposed approach and estimated timeline 3. Approximate cost breakdown by phase 4. Which YOLO architecture you’d recommend for this use case and why 5. Your strategy for sourcing quality images across all 26 species Tags / Skills Computer Vision · YOLO · Ultralytics · Object Detection · Roboflow · CoreML · iOS Machine Learning · Dataset Annotation · Image Classification · Deep Learning · Python · PyTorch · Wildlife Detection