Models trained for PowerAim

These models are in the in-app store (Models tab, Downloadable Models) under the name “… by PowerAim”. Each one detects a single class, enemy, and was trained as YOLO11s at 640 px on public datasets from Roboflow Universe. Heads and teammates in those datasets were left out of the labels; images that showed marked teammates were left out entirely.

The minimum confidence column is where the model was most accurate on its validation images. Set AI Minimum Confidence in Settings to about that value when you use the model.

Model Min. confidence mAP50 Trained on
Modern Warfare 4 (10K) 40 % 0.91 mw4 by bo7v1, acuity-mw4-take-2 by ai-model-yhlik
Battlefield 6 (6.3K) 40 % 0.84 Battlefield 6 by bf6, Battlefield 6 players by enes, battlefield by madara
Warzone (6.5K) 50 % 0.87 warzone by radek k (public domain), cod-mw-warzone by kolly
Black Ops 6 & 7 (1.2K) 35 % 0.61 Black Ops 6 SheeksterZens by Spectre Divide, Black Ops 6 by Project Locker, black-ops-enemy-tracking by cagganite
Delta Force (4.4K) 40 % 0.62 Delta Force by yolov11, Delta Force by asd, Delta Force Hawk Ops by rust-ai-cheat
Splitgate 2 (19.6K) 45 % 0.97 giga_data_splitgate by splitgate2dataset (public domain)

The number in brackets is the number of images the model learned from. Black Ops and Delta Force had the smallest or least consistent datasets and miss more enemies than the others. The Splitgate score is probably higher than what you will see in the game, because its validation images come from the same recordings as the training images.

Licence

The datasets are licensed under CC BY 4.0 unless marked public domain above. The credits above and in each model file (ONNX metadata sources) are the attribution the licence asks for. The models were trained with Ultralytics.


PowerAim is source-available under PolyForm Noncommercial. Commercial use prohibited.

This site uses Just the Docs, a documentation theme for Jekyll.