artifacts/vg_5k/manifest.json


1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18

{
  "dataset": "ranjaykrishna/visual_genome",
  "seed": 20260728,
  "selected_nodes": 5000,
  "eligible_nodes": 107019,
  "views_per_node": 16,
  "minimum_available_regions": 32,
  "extra_closure_tiers": false,
  "nested_scaling_sizes": [
    5000
  ],
  "training_files": {
    "vision": "vision_nodes.private.jsonl",
    "text": "text_nodes.jsonl"
  },
  "evaluation_only": "ground_truth.private.jsonl",
  "leakage_note": "The source image ID and URL are required only by the vision feature preprocessor. Alignment training must consume extracted features with source metadata removed."
}