What is this?
An aim-lock assistant for the game Delta Force (三角洲行动) that watches the game feed through OBS, runs a custom YOLO-omni object-detection model trained to recognize in-game characters, and then drives simulated mouse/keyboard input for smooth tracking and recoil suppression. It ships a PC (Python) version and an Android APK version, plus an auto-fishing module.
An aim-lock assistant for the game Delta Force (三角洲行动) that watches the game feed through OBS, runs a custom YOLO-omni object-detection model trained to recognize in-game characters, and then drives simulated mouse/keyboard input for smooth tracking and recoil suppression. It ships a PC (Python) version and an Android APK version, plus an auto-fishing module.Does it hook into the game process, or just watch the screen?It never touches the game's memory — it only reads the screen via OBS capture and sends simulated input through the public SendInput API.
It never touches the game's memory — it only reads the screen via OBS capture and sends simulated input through the public SendInput API.How do I get it running?Star the repo, Fork it to your own GitHub account, then download the fork as a ZIP from Code → Download ZIP; the desktop code lives in desktop/ and the mobile scripts live in mobile/, each with its own README to follow.
Star the repo, Fork it to your own GitHub account, then download the fork as a ZIP from Code → Download ZIP; the desktop code lives in desktop/ and the mobile scripts live in mobile/, each with its own README to follow.Do I need OBS, Python and a specific model file?Yes — the PC side needs OBS, Python and OpenCV, and both the PC and Android versions require you to separately download the YOLO-omni pretrained weight file before anything will run.
Yes — the PC side needs OBS, Python and OpenCV, and both the PC and Android versions require you to separately download the YOLO-omni pretrained weight file before anything will run.Is this actually safe to use in a real match?The author explicitly says this version is meant only for technical study and comparison, not for use in real games, and that every new game season can break the detection logic until the model is patched.
The author explicitly says this version is meant only for technical study and comparison, not for use in real games, and that every new game season can break the detection logic until the model is patched.How do I get the Android version?The mobile backup repo only ships a download script with no hardcoded APK address — you must supply your own Hugging Face repo and file path, then check the publisher's checksum and signature before installing.
The mobile backup repo only ships a download script with no hardcoded APK address — you must supply your own Hugging Face repo and file path, then check the publisher's checksum and signature before installing.What is the new auto-fishing module?Added in V5.0.0 as core/autofish.py, it uses visual-center detection, multi-frame voting and skeleton-point localization to find the fishing float, then drives simulated cast/reel input in water areas like Tidal Prison and the AZ3 reservoir.
Added in V5.0.0 as core/autofish.py, it uses visual-center detection, multi-frame voting and skeleton-point localization to find the fishing float, then drives simulated cast/reel input in water areas like Tidal Prison and the AZ3 reservoir.Which games does the detection model target?It's built around Delta Force (三角洲行动), with the README also naming Call of Duty and PUBG-style games as other titles where the Game2Real domain adaptation lets the model reliably tag game characters as 'person'.
It's built around Delta Force (三角洲行动), with the README also naming Call of Duty and PUBG-style games as other titles where the Game2Real domain adaptation lets the model reliably tag game characters as 'person'.Why does the project release a new major version every season?The game's maps, operators and anti-cheat (ACE) change every season, so old detection models and evasion tricks stop working — each release (e.g. V5.0.0 for the S11 'Echo' season) retrains and patches the model and anti-detection logic for the new content.
The game's maps, operators and anti-cheat (ACE) change every season, so old detection models and evasion tricks stop working — each release (e.g. V5.0.0 for the S11 'Echo' season) retrains and patches the model and anti-detection logic for the new content.What happened in earlier versions that got fixed?V3 mistakenly flagged a container hazmat suit on the AZ3 nuclear-plant map as a real player; V4/V5 added a dedicated contour filter and non-human labeling so that costume is excluded from targeting.
V3 mistakenly flagged a container hazmat suit on the AZ3 nuclear-plant map as a real player; V4/V5 added a dedicated contour filter and non-human labeling so that costume is excluded from targeting.What are similar projects?March7thAssistant, MaaEnd, X-AnyLabeling, opencv, Ai-Learn, kartpad, iPhoneMirror, Awesome-pytorch-list
March7thAssistant, MaaEnd, X-AnyLabeling, opencv, Ai-Learn, kartpad, iPhoneMirror, Awesome-pytorch-listWhich topics is it in?hack, cv, opencv, object-detection, hacking-tool, android-app, game, gaming, mobile-app, computer-vision, china, deep-learning
hack, cv, opencv, object-detection, hacking-tool, android-app, game, gaming, mobile-app, computer-vision, china, deep-learning