config.json

{
  "schema_version": "1.0",
  "model": {
    "framework": "pytorch",
    "class": "ResNetSimCLR",
    "backbone": "resnet18",
    "input_shape": [
      3,
      224,
      224
    ],
    "backbone_representation_dim": 512,
    "projection_dim": 128,
    "projection_head": [
      {
        "type": "Linear",
        "in_features": 512,
        "out_features": 512
      },
      {
        "type": "ReLU"
      },
      {
        "type": "Linear",
        "in_features": 512,
        "out_features": 128
      }
    ],
    "name": "VEDB-SimCLR-ResNet18-Baseline"
  },
  "checkpoint": {
    "format": "pytorch_checkpoint",
    "epoch": 120,
    "state_dict_key": "state_dict",
    "state_dict_entries": 124,
    "contains_optimizer_state": true,
    "strict_load_verified": true,
    "filename": "simclr_resnet18_baseline_epoch120.pth.tar"
  },
  "data": {
    "dataset": "Visual Experience Dataset (VEDB)",
    "retained_sessions": 514,
    "total_rows": 433564,
    "splits": {
      "train": {
        "sessions": 455,
        "rows": 377462
      },
      "validation": {
        "sessions": 28,
        "rows": 26026
      },
      "test": {
        "sessions": 31,
        "rows": 30076
      }
    },
    "simclr_pretraining_split": [
      "train"
    ],
    "simclr_pretraining_rows": 377462,
    "split_unit": "video_session",
    "require_ok": true
  },
  "training": {
    "epochs": 120,
    "batch_size": 512,
    "optimizer": {
      "name": "Adam",
      "learning_rate": 0.0006,
      "weight_decay": 0.0001
    },
    "objective": {
      "name": "NT-Xent",
      "implementation": "InfoNCE-style logits with CrossEntropyLoss",
      "temperature": 0.07,
      "n_views": 2,
      "feature_normalization": "L2",
      "similarity": "dot_product"
    },
    "mixed_precision": {
      "enabled": true,
      "implementation": "torch.cuda.amp",
      "gradient_scaling": true
    },
    "scheduler": {
      "name": "CosineAnnealingLR",
      "t_max": "len(train_loader)",
      "eta_min": 0.0,
      "step_unit": "epoch",
      "step_start_epoch_index": 10,
      "implementation_note": "T_max was defined as len(train_loader) while scheduler.step() was called once per epoch."
    }
  },
  "runtime_augmentations": {
    "backend": "kornia",
    "device": "gpu",
    "random_resized_crop": {
      "size": [
        224,
        224
      ],
      "scale": [
        0.08,
        1.0
      ],
      "ratio": [
        0.75,
        1.3333333333333333
      ],
      "p": 1.0
    },
    "random_horizontal_flip": {
      "p": 0.5
    },
    "color_jitter": {
      "brightness": 0.8,
      "contrast": 0.8,
      "saturation": 0.8,
      "hue": 0.2,
      "p": 0.8
    },
    "random_grayscale": {
      "p": 0.2
    },
    "gaussian_blur": {
      "kernel_size": [
        23,
        23
      ],
      "sigma": [
        0.1,
        2.0
      ],
      "p": 1.0
    },
    "normalization": {
      "mean": [
        0.485,
        0.456,
        0.406
      ],
      "std": [
        0.229,
        0.224,
        0.225
      ]
    }
  },
  "dataloader": {
    "workers": 8,
    "multiprocessing_context": "spawn",
    "pin_memory": true,
    "persistent_workers": true,
    "prefetch_factor": 4,
    "timeout_seconds": 0,
    "shuffle": true,
    "drop_last": true
  },
  "sample_error_handling": {
    "max_bad_sample_retries": 10,
    "fallback_rgb": [
      127,
      127,
      127
    ]
  },
  "reproducibility": {
    "torch_manual_seed": 0,
    "cli_seed": null,
    "fully_deterministic": false
  },
  "publication": {
    "title": "Eccentricity-Constrained CNN Training Reveals Adaptive Information Coding Around the Visual Field",
    "authors": [
      "Dylan M. Diaz",
      "Margaret M. Henderson"
    ],
    "year": 2026,
    "doi": "10.32470/0416gfsq",
    "arxiv": "2607.19316"
  },
  "provenance": {
    "technical_metadata_verified_against": [
      "released_checkpoint",
      "original_training_code",
      "cluster_launch_script",
      "offline_preprocessing_specification"
    ],
    "precedence_note": "For technical details of the released model artifact, this config and the associated Hugging Face model card reflect post-publication verification against the released checkpoint and original implementation. Where a technical implementation detail differs from the paper, the verified model artifact documentation should be used to reproduce the released checkpoint."
  },
  "condition": {
    "name": "baseline",
    "description": "Full-field VEDB imagery without an eccentricity-specific restriction.",
    "gaze_contingent": false,
    "offline_preprocessing": {
      "common": [
        "decode sampled video frame",
        "convert to RGB",
        "bicubic resize to 256 px",
        "center crop to 224 x 224"
      ],
      "condition_specific": []
    }
  }
}