🛡️ SecurePixel: Hunter Model (Scratch 512px — Asymmetric 3:1)

Model forensik Hunter berbasis ConvNeXtV2-Tiny yang dilatih murni dari awal (from scratch / zero pretraining) dengan mekanisme penalti asimetris untuk meminimalkan False Negative pada citra sintetis AI.

⚙️ Spesifikasi Training:

Arsitektur: convnextv2_tiny (Global Response Normalization)
Tipe Pelatihan: From Scratch (FP32 Murni, No Quantization, No FP16)
Input: Pure 512x512 non-overlapping crop (No resize)
Loss Function: Asymmetric Weighted Cross-Entropy:
- Weight Real (0): 1.0
- Weight AI (1): 3.0 (Hukuman 3x lipat jika AI lolos)
Final Train Loss: 0.2525
Running Train Accuracy: 81.25%
Dataset Parity: 1:1 Strict Parity per batch:
- Index 1 (AI): Kisi laten generator AI (Defactify, Gemini Imagen, FLUX, SDXL)
- Index 0 (Real): Derau optik sensor kamera riil (D05–D40)

Downloads last month: -; Downloads are not tracked for this model. How to track