Archive README.md
Browse files
README.md
CHANGED
|
@@ -1,3 +1,30 @@
|
|
| 1 |
# motion_example_0506: indexed tar archive
|
| 2 |
|
| 3 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
# motion_example_0506: indexed tar archive
|
| 2 |
|
| 3 |
+
Uncompressed tar shards preserving source-relative paths, including all modalities.
|
| 4 |
+
Original regular files: 1,498,378. Archive shards: 71.
|
| 5 |
+
|
| 6 |
+
Wait for `REMOTE_VERIFIED.json` before treating this upload as complete.
|
| 7 |
+
No dataset license is inferred or granted by this packaging operation.
|
| 8 |
+
|
| 9 |
+
## Read a member without extracting the dataset
|
| 10 |
+
|
| 11 |
+
Download the index and required tar shards, plus `pack_motion_example.py`:
|
| 12 |
+
|
| 13 |
+
```python
|
| 14 |
+
from pack_motion_example import PackedMotionReader
|
| 15 |
+
reader = PackedMotionReader('/path/to/dataset_tar')
|
| 16 |
+
payload = reader.read_bytes('metadata.jsonl') # use an existing indexed path
|
| 17 |
+
reader.close()
|
| 18 |
+
```
|
| 19 |
+
|
| 20 |
+
`reader.open(path)` buffers one member and supports file-object readers such as `numpy.load` / `torch.load`. Existing filename-only video decoders need an adapter.
|
| 21 |
+
|
| 22 |
+
## Restore files
|
| 23 |
+
|
| 24 |
+
Keep the tar files, `archive_index.sqlite`, `archive_index.tsv`, `SHA256SUMS`, `packing_summary.json`, and both Python scripts in one folder:
|
| 25 |
+
|
| 26 |
+
```bash
|
| 27 |
+
python unpack_motion_example.py --archive-dir /path/to/dataset_tar --output-dir /path/to/new_restored_directory --workers 8
|
| 28 |
+
```
|
| 29 |
+
|
| 30 |
+
Restoration verifies SHA256 and refuses mismatching existing files. External symlink targets are not copied; external links require explicit opt-in. Hardlinks were materialized as independent files. Ownership is not restored; special permission bits are removed and owner access is retained.
|