Publish verified YOCO reproduction checkpoints and prepared evaluation assets

.gitattributes

@@ -58,3 +58,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 # Video files - compressed
 *.mp4 filter=lfs diff=lfs merge=lfs -text
 *.webm filter=lfs diff=lfs merge=lfs -text
+avp/aligned_avp_glove_train_new.hdf5 filter=lfs diff=lfs merge=lfs -text
+avp/aligned_avp_glove_val.hdf5 filter=lfs diff=lfs merge=lfs -text
+interhand/InterHand_retargeted_30fps_biased_super_mano_fk20cm_bad_joint_marked.hdf5 filter=lfs diff=lfs merge=lfs -text

NOTICE.md

@@ -0,0 +1,21 @@
+# Dataset Notice
+
+## InterHand2.6M Derivative
+
+Source: https://mks0601.github.io/InterHand2.6M/
+
+InterHand2.6M: A Dataset and Baseline for 3D Interacting Hand Pose Estimation from a Single RGB Image. Gyeongsik Moon, Shoou-I Yu, He Wen, Takaaki Shiratori, Kyoung Mu Lee. ECCV 2020.
+
+License: Creative Commons Attribution-NonCommercial 4.0 International (CC-BY-NC 4.0), https://creativecommons.org/licenses/by-nc/4.0/.
+
+Upstream license: https://github.com/facebookresearch/InterHand2.6M/blob/main/LICENSE.
+
+Modifications by the YOCO authors: retargeted numeric poses, synthetic bias augmentation, MANO FK targets, and invalid-joint/frame masks. No raw images are included.
+
+## AVP/Glove Recordings
+
+Private research recordings supplied by the YOCO authors. No public redistribution license is granted at this time.
+
+## MANO
+
+MANO geometry/model pickle files are not included and must be obtained separately under https://mano.is.tue.mpg.de/license.html.

README.md

@@ -0,0 +1,35 @@
+---
+tags:
+- yoco
+- hand-pose
+- teleoperation
+- reproduction
+---
+
+# YOCO Prepared Reproduction Data
+
+Private prepared evaluation assets for [YOCO: You Only Calibrate Once!](https://github.com/Franklin-Zhang0/Calibration_net/tree/release/yoco-website), CoRL 2026. The companion checkpoints are in `Franklin0/YOCO`.
+
+## Files
+
+| File | Purpose |
+| --- | --- |
+| `interhand/InterHand_retargeted_30fps_biased_super_mano_fk20cm_bad_joint_marked.hdf5` | Table 1: two hand groups with 20 synthetic bias variants each. Evaluate all 20 right-hand variants using the provided validity masks. |
+| `avp/aligned_avp_glove_val.hdf5` | Table 2: paired AVP/glove validation; evaluate right-hand groups 1, 2, 3. |
+| `avp/aligned_avp_glove_train_new.hdf5` | Table 2 legacy baseline cohort: use group 0 in addition to the three validation recordings. This is not training data for the reproduction run. |
+
+These HDF5 files contain numeric hand/glove pose data and associated annotations, not raw photographs or video. Preserve all group names, frame order, 21-joint ordering, group MANO betas, and `gt_joint_mask`/`bad_frame` annotations. Calibration uses K=4 sampled from frames `[0,200)`; evaluation uses `[200,T)` with seeds 0, 1, 2.
+
+Table 2 YOCO, Ridge, last-layer, and full fine-tuning use three validation recordings. The historical no-calibration and LoRA baseline rows use four recordings with different group names. Rebuild that cohort using `scripts/prepare_table2_legacy_data.py` in the source repository; do not substitute a uniform three-recording cohort when verifying those paper rows.
+
+Use the repository's pinned asset downloader and SHA256 manifest, then follow its setup, export, and verification commands. Raw InterHand downloads are not a replacement for this prepared masked HDF5.
+
+## Source And Terms
+
+The InterHand file is derived from [InterHand2.6M](https://mks0601.github.io/InterHand2.6M/) by Gyeongsik Moon, Shoou-I Yu, He Wen, Takaaki Shiratori, and Kyoung Mu Lee, ECCV 2020, and retains the upstream [CC-BY-NC 4.0 license](https://github.com/facebookresearch/InterHand2.6M/blob/main/LICENSE). YOCO preparation adds retargeting, synthetic hand-bias variants, MANO forward-kinematics targets, and invalid-joint/frame annotations. Use is noncommercial and attribution is required.
+
+The AVP/glove recordings are private YOCO research assets. No public redistribution license is granted for these recordings at this time. Access to this private repository does not override source licenses.
+
+MANO model files (`MANO_LEFT.pkl`, `MANO_RIGHT.pkl`) are excluded. Obtain them directly under the [MANO license](https://mano.is.tue.mpg.de/license.html).
+
+This bundle is the prepared evaluation subset needed for offline result reproduction, not the full training corpus.

asset_manifest.json

@@ -0,0 +1,28 @@
+{
+  "files": [
+    {
+      "repository": "datasets",
+      "path": "interhand/InterHand_retargeted_30fps_biased_super_mano_fk20cm_bad_joint_marked.hdf5",
+      "local_path": "datasets/interhand/InterHand_retargeted_30fps_biased_super_mano_fk20cm_bad_joint_marked.hdf5",
+      "bytes": 209115976,
+      "sha256": "b9dd84f8bdce4caa60c59e7bf070bd55724969069185a0074813887b26b9fa29",
+      "optional": false
+    },
+    {
+      "repository": "datasets",
+      "path": "avp/aligned_avp_glove_val.hdf5",
+      "local_path": "datasets/avp/aligned_avp_glove_val.hdf5",
+      "bytes": 8148896,
+      "sha256": "918952ba7596666c7a3ca244610e1e8e219cd66c3b457c1956fa6711b0257937",
+      "optional": false
+    },
+    {
+      "repository": "datasets",
+      "path": "avp/aligned_avp_glove_train_new.hdf5",
+      "local_path": "datasets/avp/aligned_avp_glove_train_new.hdf5",
+      "bytes": 8042816,
+      "sha256": "bd2163402dae2fbeb362fea9483f2b7dad576f1f7f8ba5cc32734ce745e480d5",
+      "optional": false
+    }
+  ]
+}

avp/aligned_avp_glove_train_new.hdf5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:bd2163402dae2fbeb362fea9483f2b7dad576f1f7f8ba5cc32734ce745e480d5
+size 8042816

avp/aligned_avp_glove_val.hdf5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:918952ba7596666c7a3ca244610e1e8e219cd66c3b457c1956fa6711b0257937
+size 8148896

interhand/InterHand_retargeted_30fps_biased_super_mano_fk20cm_bad_joint_marked.hdf5

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b9dd84f8bdce4caa60c59e7bf070bd55724969069185a0074813887b26b9fa29
+size 209115976