yuzhench commited on
Commit
3f0d9ec
·
verified ·
1 Parent(s): f4fa44e

Upload README.md with huggingface_hub

Browse files
Files changed (1) hide show
  1. README.md +39 -58
README.md CHANGED
@@ -20,8 +20,8 @@ VITRA-style hand episodes for **HOT3D**, with per-hand instructions and paraphra
20
 
21
  | | |
22
  |---|---|
23
- | episodes | **18,590** |
24
- | training samples (`index_frame_pair` rows) | **780,678** |
25
  | annotation | MANO pose + world/camera joints + per-frame extrinsics |
26
  | text | one instruction per episode + 1.96 paraphrases on average |
27
  | source frame rate | 30 fps |
@@ -137,69 +137,50 @@ made exactly this mistake in an earlier render: 12 sample images collapsed pixel
137
  stored frame count matches its index rows, and no index entry points at a missing episode.
138
 
139
  ---
140
- <!-- COLLECTION_TABLE_START -->
141
- ## The collection
 
 
 
 
 
 
 
 
 
 
 
 
 
142
 
143
- Every dataset we have taken through this pipeline, with what is published today.
144
- All repos live under [`MIT-Media-Lab`](https://huggingface.co/MIT-Media-Lab) and are
145
- **annotations only — no images, no video**.
146
 
 
 
 
 
 
 
 
 
 
 
 
147
  | dataset | episodes | training samples | our contribution | size | HF |
148
  |---|---:|---:|---|---:|---|
149
  | **EPIC-KITCHENS-100** | 149,570 | 4,019,534 | episodes + text | 8.70 GB | [`epic30-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/epic30-annotations-v1) |
150
- | **EgoTouch** | 111,159 | 3,687,389 | episodes + text + **tactile** | 20.12 GB | [`egotouch-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/egotouch-annotations-v1) |
151
  | **GigaHands** | 70,486 | 2,266,087 | episodes + text | 2.89 GB | [`gigahands-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/gigahands-annotations-v1) |
152
  | **Ego-Exo4D** | 67,051 | 1,757,474 | text only | 4.09 GB | [`egoexo4d-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/egoexo4d-annotations-v1) |
153
- | **Something-Something V2** | 52,706 | 1,124,722 | text only | 4.63 GB | [`ssv2-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/ssv2-annotations-v1) |
154
- | **OakInk2** | 28,264 | 1,371,721 | episodes + text | 1.92 GB | [`oakink2-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/oakink2-annotations-v1) |
155
  | **TACO** | 23,757 | 736,136 | episodes + text | 1.34 GB | [`taco-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/taco-annotations-v1) |
156
- | **HOT3D** | 18,590 | 780,678 | episodes + text | 1.72 GB | [`hot3d-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/hot3d-annotations-v1) |
157
- | **Tachin** | 2,422 | 97,638 | episodes + text + **tactile** | 4.13 GB | [`tachin-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/tachin-annotations-v1) |
158
  | **ARCTIC** | 12,610 | 425,796 | episodes + text | 0.85 GB | [`arctic-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/arctic-annotations-v1) |
159
- | **H2O** | 5,696 | 196,941 | episodes + text | 0.40 GB | [`h2o-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/h2o-annotations-v1) |
160
- | **total** | **542,311** | **16,464,116** | | **50.7 GB** | |
161
-
162
- **episodes** = entries in `episode_frame_index.npz`, i.e. what a training run actually sees.
163
- **training samples** = rows of `index_frame_pair`; the row number *is* the sample id.
164
-
165
- *episodes + text* means we re-cut the source ourselves at wrist-speed minima and then wrote the
166
- instructions. *text only* means the episodes are VITRA-1M's official segmentation, used unchanged,
167
- and only the instructions are ours.
168
-
169
- Episodes whose round-1 caption came back `N/A` (no object interaction) are **not** published —
170
- they are excluded from both the archive and the index, so every episode here has a usable
171
- instruction. That is why the published counts are below the totals we cut:
172
-
173
- | | episodes on disk | published | dropped as `N/A` |
174
- |---|---:|---:|---:|
175
- | EPIC-KITCHENS-100 | 151,502 | 149,570 | 1,932 (1.3%) |
176
- | EgoTouch | 147,386 | 111,159 | 36,227 (24.6%) |
177
- | GigaHands | 92,365 | 70,486 | 21,879 (23.7%) |
178
- | Ego-Exo4D | 67,051 | 67,051 | 0 |
179
- | Something-Something V2 | 52,706 | 52,706 | 0 |
180
- | OakInk2 | 37,692 | 28,264 | 9,427 (25.0%) |
181
- | TACO | 26,454 | 23,757 | 2,697 (10.2%) |
182
- | HOT3D | 22,578 | 18,590 | 3,988 (17.7%) |
183
- | Tachin | 3,019 | 2,422 | 597 (19.8%) |
184
- | ARCTIC | 14,821 | 12,610 | 2,211 (14.9%) |
185
- | H2O | 7,792 | 5,696 | 2,096 (26.9%) |
186
-
187
- `ssv2` and `egoexo4d` are 0 because VITRA-1M already dropped `N/A` upstream — their episodes
188
- are the official segmentation, so there was nothing left for us to drop. Their on-disk counts are
189
- slightly below VITRA-1M's published index (52,718 and 67,053) because round 2 marked a handful of
190
- sentences `unusable` and we deleted those episodes: **12** from `ssv2`, **2** from `egoexo4d`.
191
-
192
-
193
- ### DexYCB was removed
194
-
195
- **DexYCB was removed from this collection on 2026-08-30.** It is captured by 8 fixed RealSense
196
- cameras around a table; its own `camera.role` field reads `allocentric` on all 15,878 episodes.
197
- Unlike OakInk2, which ships an egocentric view alongside three allocentric ones, DexYCB has no
198
- head-mounted camera at all, so there was nothing to filter down to.
199
-
200
- ### Not published yet
201
-
202
- | dataset | episodes cut | where it stands |
203
- |---|---:|---|
204
- | HOI4D | — | source converted by a colleague; not re-cut |
205
  <!-- COLLECTION_TABLE_END -->
 
20
 
21
  | | |
22
  |---|---|
23
+ | episodes | **18,805** |
24
+ | training samples (`index_frame_pair` rows) | **619,680** |
25
  | annotation | MANO pose + world/camera joints + per-frame extrinsics |
26
  | text | one instruction per episode + 1.96 paraphrases on average |
27
  | source frame rate | 30 fps |
 
137
  stored frame count matches its index rows, and no index entry points at a missing episode.
138
 
139
  ---
140
+ ## Revision — 2026-09-08 (re-cut)
141
+
142
+ **This release replaces the previous one. The previous episodes contained invalid frames
143
+ and should not be used.**
144
+
145
+ A defect in our episode-cutting step let frames with `kept_frames == False`
146
+ (invalid hand pose — all-zero or NaN wrist coordinates) stay inside published episodes.
147
+ The validity mask was only used to keep a cut point from landing on an invalid frame; it did
148
+ not constrain what a segment contained. Worse, a run of invalid frames could suppress cutting
149
+ altogether, so the gap was swallowed into one long segment instead of being excluded.
150
+
151
+ Every episode here is now built from a run of consecutive valid frames, so `kept_frames` is
152
+ all-`True` by construction — verified over the whole collection: **568,369 episodes /
153
+ 16.5 M frames, zero `kept_frames == False`**. Both instruction rounds were regenerated for the
154
+ new segmentation.
155
 
156
+ Segment counts and episode ids therefore changed, and the index and splits were rebuilt:
 
 
157
 
158
+ | | previous | this release |
159
+ |---|---:|---:|
160
+ | episodes on disk | 22,578 | 22,964 |
161
+ | episodes published (with an instruction) | 18,590 | 18,805 |
162
+ | training samples | 780,678 | 619,680 |
163
+
164
+ Splits are `video_test` (test owns whole videos disjoint from train/val; train and val share
165
+ the remaining videos and are split at the episode level), balanced on **frames** at 90/5/5,
166
+ seed 1.
167
+
168
+ <!-- COLLECTION_TABLE_START -->
169
  | dataset | episodes | training samples | our contribution | size | HF |
170
  |---|---:|---:|---|---:|---|
171
  | **EPIC-KITCHENS-100** | 149,570 | 4,019,534 | episodes + text | 8.70 GB | [`epic30-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/epic30-annotations-v1) |
172
+ | **EgoTouch** | 107,364 | 3,123,675 | episodes + text + **tactile** | 17.02 GB | [`egotouch-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/egotouch-annotations-v1) |
173
  | **GigaHands** | 70,486 | 2,266,087 | episodes + text | 2.89 GB | [`gigahands-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/gigahands-annotations-v1) |
174
  | **Ego-Exo4D** | 67,051 | 1,757,474 | text only | 4.09 GB | [`egoexo4d-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/egoexo4d-annotations-v1) |
175
+ | **OakInk2** | 29,058 | 1,052,924 | episodes + text | 1.56 GB | [`oakink2-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/oakink2-annotations-v1) |
 
176
  | **TACO** | 23,757 | 736,136 | episodes + text | 1.34 GB | [`taco-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/taco-annotations-v1) |
177
+ | **HOT3D** | 18,805 | 619,680 | episodes + text | 1.51 GB | [`hot3d-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/hot3d-annotations-v1) |
 
178
  | **ARCTIC** | 12,610 | 425,796 | episodes + text | 0.85 GB | [`arctic-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/arctic-annotations-v1) |
179
+ | **H2O** | 5,845 | 200,332 | episodes + text | 0.40 GB | [`h2o-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/h2o-annotations-v1) |
180
+ | **Tachin** | 2,669 | 79,584 | episodes + text + **tactile** | 3.38 GB | [`tachin-annotations-v1`](https://huggingface.co/datasets/MIT-Media-Lab/tachin-annotations-v1) |
181
+ | **total** | **487,215** | **14,281,222** | | **41.7 GB** | |
182
+
183
+ > **Something-Something V2 was dropped** from the collection (12 fps against
184
+ > 30 fps everywhere else, so a 16-step action chunk spans 1.33 s instead of 0.53 s).
185
+ > The repository still exists but should not be used.
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
186
  <!-- COLLECTION_TABLE_END -->