{
  "schema_version": 1,
  "model": {
    "name": "RAESR released decoder (Stage 3, seed 42, step-600 EMA weights)",
    "inference_protocol": "bicubic input resize to 512px, one forward pass, then mean-only color matching (paper protocol)",
    "architecture": {
      "encoder": "native DINOv3 ViT-L/16, layers 1..23 (0-indexed; layers 2-24 in the paper) plus final-layer spatial-mean broadcast",
      "decoder": "ViT-XL RoPE",
      "resolution": 512,
      "decoder_patch_size": 16,
      "rope_theta": 100.0,
      "scale_factor": 1,
      "encoder_cls_mode": "none"
    }
  },
  "artifacts": {
    "decoder": {
      "repo_id": "raesrreseach/sr_decoder_iclr",
      "revision": "main",
      "filename": "ema_decoder.pt",
      "bytes": 1661590309,
      "sha256": "2e4de5df246172f067e0437f2346d17ca719a8dd12146624df0042983c580cb3"
    },
    "raesr_d_frontend": {
      "repo_id": "raesrreseach/sr_decoder_iclr",
      "revision": "main",
      "filename": "weights/raesr_d_frontend.pt",
      "bytes": 231182271,
      "sha256": "72f970fc93667b7c9165a0a71522bb6f4f5e0a79899102c67058cbdcfe18f7e7"
    },
    "dinov3_vitl16": {
      "source_requirement": "native facebookresearch/dinov3 checkout pinned by the runtime provenance manifest",
      "filename": "dinov3_vitl16_pretrain_lvd1689m-8aa4cbdd.pth",
      "checkpoint_suffix": "8aa4cbdd"
    },
    "dinov3_vits16plus": {
      "source_requirement": "native facebookresearch/dinov3 checkout pinned by the runtime provenance manifest",
      "filename": "dinov3_vits16plus_pretrain_lvd1689m-4057cbaa.pth",
      "checkpoint_suffix": "4057cbaa"
    }
  }
}