Update README.md
Browse files
README.md
CHANGED
|
@@ -22,7 +22,7 @@ All samples are stored in a Hugging Face-compatible parquet file at `data/<split
|
|
| 22 |
ds = load_dataset('path/to/this/folder', split='train')
|
| 23 |
sample = ds[0]
|
| 24 |
``
|
| 25 |
-
`sample["preview"]` loads directly as a PIL image; `sample["raw_image_path"]` points to the higher-quality file
|
| 26 |
3. Use the metadata to drive evaluation logic, e.g., compute city-level accuracy, filter by `data_type`, or inspect specific regions.
|
| 27 |
|
| 28 |
## Notes
|
|
|
|
| 22 |
ds = load_dataset('path/to/this/folder', split='train')
|
| 23 |
sample = ds[0]
|
| 24 |
``
|
| 25 |
+
`sample["preview"]` loads directly as a compressed PIL image; `sample["raw_image_path"]` points to the higher-quality file for inference.
|
| 26 |
3. Use the metadata to drive evaluation logic, e.g., compute city-level accuracy, filter by `data_type`, or inspect specific regions.
|
| 27 |
|
| 28 |
## Notes
|