epic30-annotations-v1

Annotations only — no images, no video.

VITRA-style hand episodes for EPIC-KITCHENS-100, with per-hand instructions and paraphrases.

episodes	149,570
training samples (index_frame_pair rows)	4,019,534
annotation	MANO pose + world/camera joints + per-frame extrinsics
text	one instruction per episode + 1.88 paraphrases on average
images / video	not included — see Getting the frames below

What we did

Episodes are ours. The source release ships either raw video or differently-segmented clips, so we re-cut it with VITRA's method — speed minima of the 3D wrist in world space:

gaussian smooth (sigma=1.0) -> local speed minima in a fixed window (win=15, i.e. 0.5 s
at 30 Hz) -> merge runs shorter than min_seg=16 -> pad 2 frames on each end

sigma and win are quantities in time, converted per source frame rate. Left and right hands are cut independently, with the other hand's motion ignored.

Instructions are ours. Two rounds, both with Qwen3.5-122B-A10B-FP8: round 1 captions 8 frames per episode with the palm's future trajectory drawn on them; round 2 checks the sentence belongs to that hand, strips same-hand references ("Rinse the right hand." -> "Rinse the hand.", because training already prepends Left hand: ... Right hand: ...), and writes 1-3 paraphrases.

Episodes with no instruction are not included. Round 1 returns N/A when an episode shows no object interaction. Those episodes are excluded from both the archive and the index, so every episode here has a usable instruction.

Splits

train.txt / val.txt / test.txt under splits/ list episode ids, one per line. splits/test_videos.txt lists the videos the test set owns.

	videos	episodes	frames	share
train	548	134,522	3,615,354	89.94%
val	494	7,527	200,835	5.00%
test	32	7,521	203,345	5.06%

test owns its videos outright — none of its 32 videos appears in train or val. train and val share videos: within each non-test video, episodes are split between the two. So val measures fit on unseen clips from seen videos (use it for early stopping and hyper-parameter choice); test is the only held-out set.

Grouping is by video, not by participant. EPIC participants each film many videos in their own kitchen, so a participant whose videos land in test also has videos in train — the test set contains no unseen kitchen. It measures generalisation across clips and videos within known environments, not across environments.

Note on P18_09

P18_09 is the only 90 fps video in EPIC-KITCHENS-100 (GoPro Hero7 1080p90; every other video is 59.94 / 50 / 29.97 fps). Its annotations are sampled every 3rd frame rather than every 2nd, which an earlier version of our cutter treated as discontinuous — it produced 0 episodes for that video. Fixed here: 155 episodes.

Files

epic30.tar                 ->  Annotation/epic30/episodic_annotations/*.npy
episode_frame_index.npz        index_frame_pair (N,2) uint32 + index_to_episode_id (E,)

index_frame_pair row number is the sample id: row r = (episode ordinal, frame within that episode). len(index_frame_pair) is the size of the training set.

import numpy as np
# tar -xf epic30.tar
z = np.load("episode_frame_index.npz", allow_pickle=True)
ep_slot, frame_id = z["index_frame_pair"][sample_id]
eid = str(z["index_to_episode_id"][ep_slot])
d = np.load(f"Annotation/epic30/episodic_annotations/{eid}.npy", allow_pickle=True).item()
rgb_frame_id = int(d["video_decode_frame"][frame_id])

Each .npy is a dict with video_name, video_decode_frame, intrinsics, per-frame extrinsics (world->camera), anno_type (which hand this episode is for), text, text_rephrase, and a left/right dict holding beta, hand_pose, global_orient_worldspace, transl_worldspace, joints_worldspace, kept_frames. text[hand] = [(sentence, (0, T))] and text_rephrase[hand] = [([paraphrases...], (0, T))].

Getting the frames

video_decode_frame indexes the source video, which we do not redistribute. Get it from EPIC-KITCHENS-100 — https://epic-kitchens.github.io/, then decode by index (we use decord; a self-maintained sequential counter drifts silently if the decoder ever skips a frame).

Known limitations

• Paraphrase count averages 1.88, not a fixed number. Past 3 the model starts inventing; a sentence with no prepositional phrase honestly supports only one or two.
• Verified: the index lists exactly the episodes that have an instruction, every episode's stored frame count matches its index rows, and no index entry points at a missing episode.

---

The collection

Every dataset we have taken through this pipeline, with what is published today. All repos live under MIT-Media-Lab and are annotations only — no images, no video.

dataset	episodes	training samples	our contribution	size	HF
EPIC-KITCHENS-100	149,570	4,019,534	episodes + text	8.70 GB	epic30-annotations-v1
EgoTouch	111,159	3,687,389	episodes + text + tactile	20.12 GB	egotouch-annotations-v1
GigaHands	70,486	2,266,087	episodes + text	2.89 GB	gigahands-annotations-v1
Ego-Exo4D	67,051	1,757,474	text only	4.09 GB	egoexo4d-annotations-v1
Something-Something V2	52,706	1,124,722	text only	4.63 GB	ssv2-annotations-v1
OakInk2	28,264	1,371,721	episodes + text	1.92 GB	oakink2-annotations-v1
TACO	23,757	736,136	episodes + text	1.34 GB	taco-annotations-v1
ARCTIC	12,610	425,796	episodes + text	0.85 GB	arctic-annotations-v1
H2O	5,696	196,941	episodes + text	0.40 GB	h2o-annotations-v1
total	521,299	15,585,800		44.9 GB

episodes = entries in episode_frame_index.npz, i.e. what a training run actually sees. training samples = rows of index_frame_pair; the row number is the sample id.

episodes + text means we re-cut the source ourselves at wrist-speed minima and then wrote the instructions. text only means the episodes are VITRA-1M's official segmentation, used unchanged, and only the instructions are ours.

Episodes whose round-1 caption came back N/A (no object interaction) are not published — they are excluded from both the archive and the index, so every episode here has a usable instruction. That is why the published counts are below the totals we cut:

	episodes on disk	published	dropped as N/A
EPIC-KITCHENS-100	151,502	149,570	1,932 (1.3%)
EgoTouch	147,386	111,159	36,227 (24.6%)
GigaHands	92,365	70,486	21,879 (23.7%)
Ego-Exo4D	67,051	67,051	0
Something-Something V2	52,706	52,706	0
OakInk2	37,692	28,264	9,427 (25.0%)
TACO	26,454	23,757	2,697 (10.2%)
ARCTIC	14,821	12,610	2,211 (14.9%)
H2O	7,792	5,696	2,096 (26.9%)

ssv2 and egoexo4d are 0 because VITRA-1M already dropped N/A upstream — their episodes are the official segmentation, so there was nothing left for us to drop. Their on-disk counts are slightly below VITRA-1M's published index (52,718 and 67,053) because round 2 marked a handful of sentences unusable and we deleted those episodes: 12 from ssv2, 2 from egoexo4d.

DexYCB was removed

DexYCB was removed from this collection on 2026-08-30. It is captured by 8 fixed RealSense cameras around a table; its own camera.role field reads allocentric on all 15,878 episodes. Unlike OakInk2, which ships an egocentric view alongside three allocentric ones, DexYCB has no head-mounted camera at all, so there was nothing to filter down to.

Not published yet

dataset	episodes cut	where it stands
HOI4D	—	source converted by a colleague; not re-cut
HOT3D	—	source converted by a colleague; not re-cut