AI 1.0.17

AI 1.0.17

• Restore the original RookieAI YOLOv8s Apex model used before the local fine-tune.
• Remove the local fine-tuned model weights, exports and metadata from the current source tree and packaged runtime.
• Keep the existing configuration values and the other changes from 1.0.16.

The restored engine is byte-identical to the v1.0.13 backup (SHA-256:
1a20af2bae6aa82257c4a3ce66419647a4e38cd0e3b1a79f72533525c81e73b5).

Validation: native Debug/Release builds, five JavaScript tests, runtime self-test,
packaged/installed model hashes and model selection, installer exit code 0, and
CUDA inference using the installed original TensorRT engine. Existing settings
values are preserved; the retired experimental suppression field is removed by
the configuration migration already introduced in 1.0.16. In-game detection
and aiming were not tested.

Installer SHA-256: 8f78a54b44b3b0d32c0bcd98213cb8a21be58def62d8c1726bf084f877fb0c15.

AI 1.0.15

AI 1.0.15

• Add an experimental switch under Settings → Advanced settings to suppress physical mouse movement away from the selected aim point while AI aim is active. It is off by default and changes apply live. The input core must be running.
• Compare each physical movement with the direction to the aim point. Opposing movement is consumed; movement toward the point or perpendicular to it passes through. Mouse buttons and the wheel remain available.
• Release the aim trigger to move freely. The gate also releases on focus loss, missing or stale target data, shutdown, and inside a 2-pixel dead zone around the point.
• Move Mouse connection / Reconnect mouse from the home page into Advanced settings, including status and error feedback.

Validation: native Debug/Release builds; native direction/expiry tests; Python-to-C++ shared-memory integration; AI output-loop lifecycle checks with a recording driver (no mouse injection); configuration persistence and frontend checks. Actual in-game feel requires user testing.

Installer SHA-256: 59b531fe2af0e590a46de893956a4813f39b03cd1c4827b334a8b14bdc569692

AI 1.0.14

AI 1.0.14

• 換用第一輪 RookieAI YOLOv8s Apex 續訓模型。
• 全新設定的信心門檻為 0.5，既有使用者可在 GUI 自行調整。
• 包含模型來源、資料署名與模型檔案雜湊。

驗證

• Native Debug/Release build 與 AI 設定測試通過。
• TensorRT 匯出在 75 張裁切畫面中與原 .pt 的偵測數量一致；最低框 IoU 0.954。
• 本機 NSIS 安裝退出碼 0，安裝後的版本、9 個 runtime 檔案及模型雜湊核對通過。
• 來源程式確認新模型與 CUDA ready；未完成安裝版 GUI 啟動及即時遊戲互動測試。

模型成果與限制

1,180 張本輪訓練留出圖片、conf 0.25 / IoU 0.5：一般大小敵人召回 69.1% → 91.5%，24–48 px 30.0% → 66.7%，<24 px 1/15 → 3/15。
新模型 conf 0.5 的整體 precision 為 81.1%、recall 為 84.4%；最小人物樣本仍少，仍有背景誤判。

模型與資料署名

• 原權重：Passer1072/RookieAI_yolov8。
• PSImera/apex_enemy_detect：CC BY-NC 4.0。
• enemy-allies-tags v2：CC BY 4.0。
• Ultralytics 元件及模型授權。

這份模型提供非商業使用；完整通知位於安裝包 resources/ApexVision/MODEL_NOTICES.md。
引擎以 RTX 3070 / TensorRT 11.2.1.2、batch 1、384×640、FP16 匯出。

安裝檔 SHA-256

de798046c90c13a0047ca4dcf8fdb1ab973661fd06b852875875ad9729a77ed2

AI 1.0.12

AI 1.0.12 removes the separate mouse sensitivity and ADS multiplier controls from AI aiming. Existing settings are migrated to one effective yaw-per-count value, preserving the prior mouse movement scale. The existing 1.0.11 ByteTrack service and TensorRT model remain bundled.

The AI service still needs a compatible Python environment with the packages in ApexVision/requirements.txt. The bundled TensorRT engine is GPU-specific.

Private source: SmallSideOUO/AI-Apex commit 65817675df3e4177a80394d098b28e108ff9f88f.

AI 1.0.11

AI 1.0.11 includes ByteTrack target tracking in the ApexVision service. The installer contains the updated AI scripts and TensorRT model, and the installed app runs these bundled files.

The AI service still needs a compatible Python environment with the packages in ApexVision/requirements.txt. The bundled TensorRT engine is GPU-specific.

Private source: SmallSideOUO/AI-Apex commit 5a14f275f588c01bebd7d321f191a87f4fb8dc22.

AI v1.0.10

Changes

• Show AI detection boxes directly over the Apex window with a transparent, click-through overlay.
• Highlight the active target and keep the overlay aligned with the game client area.
• Replace the image preview in the control panel with a box overlay switch and update-rate control.

The ApexVision Python runtime and model remain separate from this installer and are included in the private AI-Apex source repository.

AI v1.0.9

Remove the unused AI right-stick output path and migrate existing AI settings to mouse-only output. Add an adjustable maximum mouse movement speed in the PID controls and clarify P, I, D, integral limit, and output rate.

AI v1.0.8

Rename the Xim input-core control to AI in all supported languages and clarify its movement and recoil features.

AI v1.0.7

固定使用原生鍵盤與滑鼠輸入，移除虛擬手把及模式切換。

• Mouse5 不再切換模式；保留 Mouse5 + Mouse4 + 左右鍵的壓槍觸發組合。
• AI 瞄準改為滑鼠輸出，既有設定會在新版程式讀取時轉換。
• 移除介面上的手把映射與阻擋設定，後端啟動不再依賴 ViGEm。
• 安裝程式包含桌面 GUI 與 Xim 後端。ApexVision 的 Python 執行環境與模型需在使用 AI 功能的電腦另外設定。

Windows installer SHA-256: 59A69781A866A937E35ACDFF17B23F3BC3F522BE9F0B5CBF82D6D966B38B5E45

AI v1.0.6

修正語言切換初始化錯誤與驅動狀態停在 Check 的問題。

• 預設語言改為 English，左下語言選單會明確選取目前語言
• 完成繁中、簡中、英文、日文的介面與進階設定翻譯
• 永久授權與剩餘時間會跟著語言切換
• 驅動已啟動時正確顯示 Running
• 乾淨安裝預設使用目前設定與 MnK 模式