focalnet-human.onnx.json

{
  "format_version": 2,
  "task": "human-crop-ranking",
  "input": {
    "image": [
      1,
      3,
      256,
      256
    ],
    "boxes": [
      1,
      128,
      4
    ],
    "content": [
      1,
      4
    ],
    "box_coordinates": "normalized oriented source image",
    "content_coordinates": "normalized letterboxed model input",
    "channels": "RGB",
    "mean": [
      0.48500001430511475,
      0.4560000002384186,
      0.4059999883174896
    ],
    "std": [
      0.2290000021457672,
      0.2240000069141388,
      0.22499999403953552
    ],
    "resize": "bilinear-letterbox"
  },
  "output": {
    "importance": {
      "shape": [
        1,
        1,
        64,
        64
      ],
      "activation": "sigmoid already applied"
    },
    "crop_scores": {
      "shape": [
        1,
        128
      ],
      "semantics": "relative human-preference logits; compare only within one image"
    }
  },
  "model_config": {
    "backbone": "repvit_m0_9",
    "decoder_channels": 48
  },
  "rank_config": {
    "feature_channels": 16,
    "hidden_channels": 64
  },
  "parameters": 4757694,
  "ranking_parameters": 7069,
  "trained_epochs": 28,
  "checkpoint_sha256": "82c4d695310c220e5f3bcfbf5e64d315a5e826cfd77e945e61e032460014480a",
  "base_checkpoint_sha256": "d1942f0652f8ea85f75ffc0cb1bf40d7e70b38e7ad102ab0e6df5c2e07ce52cf",
  "initialize_checkpoint_sha256": "16cf17814cbeec88180d158fd6ba018355e6ca80aecc7f861d7325e295b9c91b",
  "dataset_name": "GAICD journal version",
  "dataset_fingerprint": "b99bffff76228cb2d89077a772586b9874d3879e0b4e90a323e65e719e3197a8",
  "candidate_policy": {
    "positions": 5,
    "scales": [
      1.0,
      0.9,
      0.8,
      0.7,
      0.65
    ],
    "retention_tolerance": 0.05
  },
  "onnxruntime_verification": {
    "provider": "CPUExecutionProvider",
    "graph_optimization": "ORT_DISABLE_ALL",
    "atol": 0.05,
    "rtol": 0.001
  },
  "export_max_abs_error": 0.004714310169219971
}