Update README.md
Browse files
README.md
CHANGED
|
@@ -41,24 +41,6 @@ This Hugging Face version has two loadable configs:
|
|
| 41 |
|
| 42 |
The original annotation JSONL files are also included under `original_annotations/` for legacy compatibility.
|
| 43 |
|
| 44 |
-
## Source archive note
|
| 45 |
-
|
| 46 |
-
The hosted `vcr1images.zip` object has one JPEG member whose zip CRC does not match, although the member's full byte stream can be recovered and decoded successfully:
|
| 47 |
-
|
| 48 |
-
- `vcr1images/lsmdc_1010_TITANIC/1010_TITANIC_02.33.49.825-02.33.53.701@0.jpg`
|
| 49 |
-
|
| 50 |
-
Rows whose image comes from that zip member have `source_zip_crc_mismatch=true`.
|
| 51 |
-
|
| 52 |
-
The `width` and `height` columns preserve the original metadata JSON dimensions used by boxes/segmentations and by legacy `r2c` code. `image_width` and `image_height` store the decoded JPEG dimensions. Rows where those differ have `image_metadata_dimension_mismatch=true`.
|
| 53 |
-
|
| 54 |
-
## Legacy r2c layout
|
| 55 |
-
|
| 56 |
-
The original `rowanz/r2c` code expects `train.jsonl`, `val.jsonl`, `test.jsonl`, and a `vcr1images/` directory. Use:
|
| 57 |
-
|
| 58 |
-
```bash
|
| 59 |
-
python scripts/export_r2c_layout.py <repo_id> /path/to/r2c/data
|
| 60 |
-
```
|
| 61 |
-
|
| 62 |
## License and terms
|
| 63 |
|
| 64 |
Use of VCR is governed by the original VCR/AI2 dataset license and website terms:
|
|
|
|
| 41 |
|
| 42 |
The original annotation JSONL files are also included under `original_annotations/` for legacy compatibility.
|
| 43 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 44 |
## License and terms
|
| 45 |
|
| 46 |
Use of VCR is governed by the original VCR/AI2 dataset license and website terms:
|