| # Complementary NTIRE Real images |
|
|
| This prefix contains 25,000 additional NTIRE `label=0` Real |
| images. Every image has `width > 512` and |
| `height > 512`. The original 12,607 rows remain in |
| seven immutable `train-xxxxx-of-00007.parquet` files; the 12,393 |
| `shard_4` continuation is stored in seven |
| `train-append-shard4-xxxxx-of-00007.parquet` files. |
|
|
| The append selection excludes every `sample_id` or image SHA-256 already in |
| the 50K `NTIRE/real` set or the original Complementary set. Original JPEG bytes |
| are embedded without re-encoding and every row matches profile |
| `d39db51fdcf2ed8c0f2148e77eb00b5f48d4993e8fa2d5e501a28188e1fd9f44`. |
|
|
| ```python |
| from datasets import load_dataset |
| |
| dataset = load_dataset( |
| "Edward0707/TechJam2026_Jamlai_Bench", |
| data_files={"train": "Complementary/train-*.parquet"}, |
| ) |
| ``` |
|
|
| `selection_manifest.csv` contains row-level provenance and hashes; |
| `summary.json` records both immutable generations and their validation. |
|
|
