Finalize verified WorldRide F19 rebuttal data release
Browse files
rebuttal_data/f19/RESTORE.txt
ADDED
|
@@ -0,0 +1,11 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
WorldRide F19 rebuttal data
|
| 2 |
+
|
| 3 |
+
All three packages are required. Original file contents are preserved. No raw training videos are included.
|
| 4 |
+
|
| 5 |
+
1. Verify packages with sha256sum -c SHA256SUMS (macOS: shasum -a 256 -c SHA256SUMS).
|
| 6 |
+
2. Extract worldride_f19_metadata.tar.gz into your data staging directory. It contains manifests/three_source_windows_k_complete_v1, calibration/exclusion.json, calibration/gt_conditioning_target_rules.json, and conditioning_targets/three_source_k_complete_v5.
|
| 7 |
+
3. Extract initial_latents_f19.tar.gz into the project output directory to restore output/initial_latents_f19. This includes all 16 original LMDB shards and verification metadata.
|
| 8 |
+
4. Extract annotations_f19_exact.tar.gz into your world_model/Annotation directory. Paths are relative to that Annotation root. Only manifest-referenced pose/prompt paths and existing intrinsics candidates were selected; no video files were downloaded.
|
| 9 |
+
5. Remap historical absolute /mnt/workspace/fly/reward_dmd_sanawm and /mnt/data/shared/datasets/world_model paths in your runtime configuration as needed. Do not change original manifest content before checking its source SHA256. Training code and model weights are separate prerequisites.
|
| 10 |
+
|
| 11 |
+
RELEASE.json records package hashes and source counts. annotations-receipt.tsv.gz contains per-file source OSS paths, byte sizes and SHA256.
|