Update README (license + structure)

README.md

@@ -1,5 +1,5 @@
 ---
-license: cc-by-4.0
+license: cc-by-nc-4.0
 task_categories:
 - object-detection
 tags:
@@ -17,8 +17,17 @@ Real-world hand-held capture used as the primary test set in
 [Event6D](https://github.com/mickeykang16/Event6D) (CVPR 2026). With the provided
 checkpoints, our method achieves **mean ADDS 52.79** on this set.
 
-A Prophesee event camera + Intel RealSense RGB-D camera filming 14 daily YCB-style
-objects from random angles/motions.
+Hand-held capture of 14 daily YCB-style objects from random viewpoints and motions.
+
+## Hardware
+
+| Sensor | Model | Resolution | Rate |
+|---|---|---|---|
+| RGB-D | Intel RealSense D435i | 1280 × 720 | 30 FPS |
+| Event | Prophesee IMX636 | 1280 × 720 | ≥ 5000 FPS |
+
+Calibration (intrinsics + RGB↔event extrinsic) is shipped as `0001-camchain.yaml`
+in the Kalibr camchain format.
 
 ## Layout
 
@@ -33,30 +42,57 @@ Event6D/
     ├── depth_aligned_to_color/
     ├── depth_aligned_to_event/
     ├── pose/                 # GT object poses (4, 4, 4): 4 sub-timesteps per file
-    ├── pose_orig/            # only for wine_1101/0001 — original poses before frame-shift fix
     ├── mask/
     ├── obj.txt               # object id string (e.g. 011_wine_glass)
     ├── startend.txt
     └── aligned_depth_pose.csv
 ```
 
-## What's omitted (and how to regenerate)
+## Ground-truth pose format
+
+Although RGB is captured at 30 FPS, the event camera produces a continuous stream that
+we sample at 120 Hz. To give pose labels at the finer rate, each `pose/XXXXXX.npy` stores
+**4 poses** instead of one — these are 4 evenly-spaced sub-timesteps that span the
+interval between RGB frame `XXXXXX` and the next RGB frame `XXXXXX+1`.
+
+```
+file shape: (4, 4, 4)
+              │  └──┴── 4×4 SE(3) rigid-body transform (object → camera)
+              └──── sub-timestep index 0..3 within one 30 FPS frame interval
+```
 
-To keep the dataset under ~25 GB we skipped the precomputed event representations:
-`parsed_voxel_depth*`, `parsed_voxel_recon*`, `parsed_e2vid*`, `event_viz*`, `result_viz`,
-`depth_qual*`. These are caches generated by the data reader on first use — running
-evaluation against this dataset will lazily compute and write them under your local copy,
-adding ~1 GB per sequence (~5-10 min one-time warmup per sequence).
+- `pose[0]` is co-temporal with `rgb/XXXXXX.jpg` (the start of the interval).
+- `pose[1]`, `pose[2]`, `pose[3]` are 1/4, 2/4, 3/4 of the way to `rgb/XXXXXX+1.jpg`.
 
-## Per-sequence sub-timestep convention
+Evaluators can therefore choose either:
+- **30 FPS** — use only `pose[0]` per frame, comparing against the tracker's prediction
+  at the RGB instant; or
+- **120 FPS** — use all 4 sub-timesteps, comparing against pose predictions emitted at
+  each event sub-step.
 
-Each `pose/XXXXXX.npy` has shape `(4, 4, 4)`: four temporally-spaced sub-timesteps of the
-object pose covering the interval `[T_XXXXXX, T_XXXXXX+1)`. `pose[0]` is at the same
-instant as `rgb/XXXXXX.jpg`; `pose[1..3]` are intermediate samples toward the next frame.
+## Event file timing
 
-The `wine_1101/0001` sequence originally had a one-frame offset (`pose[3]` aligned with
-RGB instead of `pose[0]`); this release ships the corrected (shifted) `pose/` and keeps
-the original under `pose_orig/` for reproducibility.
+Events are pre-packed into `parsed_events/XXXXXX.npz`. Each file holds the events that
+occurred during the **inter-frame interval ending at RGB frame `XXXXXX`** — i.e.
+the events between `rgb/(XXXXXX-1).jpg` and `rgb/XXXXXX.jpg` (a window of ≈ 1/30 s).
+
+```
+time →
+    ┌────────────────── 1 / 30 s ──────────────────┐
+    rgb/(i−1).jpg                          rgb/i.jpg
+    │                                              │
+    │←── parsed_events/i.npz  (events in interval) │
+    │                                              │
+    ├──── pose/(i−1).npy[0..3] (4 sub-timesteps) ──┤
+```
+
+So for any frame `i`:
+- `rgb/i.jpg`     — snapshot at the end of the interval
+- `parsed_events/i.npz` — events leading **up to** that snapshot
+- `pose/(i−1).npy[0..3]` — 4 GT poses across the same interval, aligned with the events
+
+Each `.npz` stores a structured array under key `data` with fields
+`(x, y, t, p)` (event pixel coords, timestamp in seconds, polarity ±1).
 
 ## Download
 
@@ -67,4 +103,17 @@ huggingface-cli download mickeykang/Event6D --repo-type dataset \
 
 ## Citation
 
-Please cite Event6D (CVPR 2026) if you use this dataset.
+```bibtex
+@inproceedings{kang2026event6d,
+  title     = {Event6D: Event-based Novel Object 6D Pose Tracking},
+  author    = {Kang, Jae-Young and
+               Cho, Hoonehee and
+               Lee, Taeyeop and
+               Kang, Minjun and
+               Wen, Bowen and
+               Kim, Youngho and
+               Yoon, Kuk-Jin},
+  booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
+  year      = {2026}
+}
+```