| Attribute | Value | |-----------|-------| | | PyTorch checkpoint ( torch.save ) | | Size on disk | ≈ 2.1 GB (fp32) – ~1.1 GB when saved with torch.save(..., _use_new_zipfile_serialization=False, pickle_protocol=4) and torch.save(..., dtype=torch.float16) | | Top‑level keys | 'encoder', 'mapper', 'generator', 'args' | | encoder | state_dict of a ResNet‑50 (BN layers stripped) | | mapper | 2‑layer MLP (512 → 512) plus LayerNorm | | generator | StyleGAN2 weights (including the new 2048‑pixel synthesis blocks) | | args | Namespace containing training hyper‑parameters, input resolution, output resolution, and a version string ( GPEN-BFR-v2.0-2048 ). | | Compatibility | Requires PyTorch ≥ 1.8 and CUDA ≥ 11.0 (or CPU‑only fallback). The checkpoint can be loaded on any device with the same architecture (ResNet‑50 + StyleGAN2). |
, offering much higher detail for close-ups and professional-grade enhancements. Primary Use Case gpen-bfr-2048.pth
This model is specifically tuned to restore severely degraded or low-quality facial images—often called "in the wild" images—improving clarity, detail, and resolution. | Attribute | Value | |-----------|-------| | |
: Fixing old, pixelated, or out-of-focus family photos. | , offering much higher detail for close-ups
gpen-bfr-2048.pth file is a high-resolution pre-trained model checkpoint for
For instance, if you are using the , you would typically place this file in the models/GFPGAN or models/GPEN directory to enable the "Face Restoration" checkbox in your interface.