loc_inspect.py

# /// script
# dependencies = ["datasets", "huggingface_hub"]
# ///
import json
from datasets import load_dataset

ds = load_dataset("biglam/loc_beyond_words")
for split in ds:
    print("=== SPLIT:", split, len(ds[split]))
    f = ds[split].features
    print("features:", list(f.keys()))
    print("features detail:", {k: str(v) for k, v in f.items()})

# look at labels in a sample row
row = ds["validation"][0]
print("sample keys:", list(row.keys()))
for k, v in row.items():
    if k != "image":
        print("  ", k, ":", v)

# gather all training classes
train = ds["train"]
classes = set()
imgs_per_class = {}
N = len(train)
for i in range(N):
    objs = train[i]["objects"]
    for el in objs.get("label", []):
        el = el if isinstance(el, int) else int(el)
        classes.add(el)
print("\nnum rows train:", N)
print("label set in train:", sorted(classes))