Datasets:
seed SCHEMA.md
Browse files
SCHEMA.md
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
# Schema & curation notes
|
| 2 |
+
|
| 3 |
+
`metadata.jsonl` — one JSON object per line, HF `imagefolder` compatible
|
| 4 |
+
(`file_name` points into `images/`).
|
| 5 |
+
|
| 6 |
+
Fields: see the dataset card. Labels are weakly-supervised (single self-report);
|
| 7 |
+
a downstream consensus / expert-review pass is expected before training use.
|
| 8 |
+
|
| 9 |
+
`leaderboard.jsonl` — game scores (separate from the training data):
|
| 10 |
+
{ "contributor", "skill_correct", "skill_total", "machine_correct",
|
| 11 |
+
"contributions", "updated" }
|