Add files using upload-large-folder tool

cb2428f verified 6 months ago

51.4 kB

	Metadata-Version: 2.2
	Name: seamless-interaction
	Version: 0.1.0
	Summary: Library for seamless interaction dataset
	Author: Meta AI \| FAIR Seamless Team
	License: Attribution-NonCommercial 4.0 International

	=======================================================================

	Creative Commons Corporation ("Creative Commons") is not a law firm and
	does not provide legal services or legal advice. Distribution of
	Creative Commons public licenses does not create a lawyer-client or
	other relationship. Creative Commons makes its licenses and related
	information available on an "as-is" basis. Creative Commons gives no
	warranties regarding its licenses, any material licensed under their
	terms and conditions, or any related information. Creative Commons
	disclaims all liability for damages resulting from their use to the
	fullest extent possible.

	Using Creative Commons Public Licenses

	Creative Commons public licenses provide a standard set of terms and
	conditions that creators and other rights holders may use to share
	original works of authorship and other material subject to copyright
	and certain other rights specified in the public license below. The
	following considerations are for informational purposes only, are not
	exhaustive, and do not form part of our licenses.

	Considerations for licensors: Our public licenses are
	intended for use by those authorized to give the public
	permission to use material in ways otherwise restricted by
	copyright and certain other rights. Our licenses are
	irrevocable. Licensors should read and understand the terms
	and conditions of the license they choose before applying it.
	Licensors should also secure all rights necessary before
	applying our licenses so that the public can reuse the
	material as expected. Licensors should clearly mark any
	material not subject to the license. This includes other CC-
	licensed material, or material used under an exception or
	limitation to copyright. More considerations for licensors:
	wiki.creativecommons.org/Considerations_for_licensors

	Considerations for the public: By using one of our public
	licenses, a licensor grants the public permission to use the
	licensed material under specified terms and conditions. If
	the licensor's permission is not necessary for any reason--for
	example, because of any applicable exception or limitation to
	copyright--then that use is not regulated by the license. Our
	licenses grant only permissions under copyright and certain
	other rights that a licensor has authority to grant. Use of
	the licensed material may still be restricted for other
	reasons, including because others have copyright or other
	rights in the material. A licensor may make special requests,
	such as asking that all changes be marked or described.
	Although not required by our licenses, you are encouraged to
	respect those requests where reasonable. More considerations
	for the public:
	wiki.creativecommons.org/Considerations_for_licensees

	=======================================================================

	Creative Commons Attribution-NonCommercial 4.0 International Public
	License

	By exercising the Licensed Rights (defined below), You accept and agree
	to be bound by the terms and conditions of this Creative Commons
	Attribution-NonCommercial 4.0 International Public License ("Public
	License"). To the extent this Public License may be interpreted as a
	contract, You are granted the Licensed Rights in consideration of Your
	acceptance of these terms and conditions, and the Licensor grants You
	such rights in consideration of benefits the Licensor receives from
	making the Licensed Material available under these terms and
	conditions.


	Section 1 -- Definitions.

	a. Adapted Material means material subject to Copyright and Similar
	Rights that is derived from or based upon the Licensed Material
	and in which the Licensed Material is translated, altered,
	arranged, transformed, or otherwise modified in a manner requiring
	permission under the Copyright and Similar Rights held by the
	Licensor. For purposes of this Public License, where the Licensed
	Material is a musical work, performance, or sound recording,
	Adapted Material is always produced where the Licensed Material is
	synched in timed relation with a moving image.

	b. Adapter's License means the license You apply to Your Copyright
	and Similar Rights in Your contributions to Adapted Material in
	accordance with the terms and conditions of this Public License.

	c. Copyright and Similar Rights means copyright and/or similar rights
	closely related to copyright including, without limitation,
	performance, broadcast, sound recording, and Sui Generis Database
	Rights, without regard to how the rights are labeled or
	categorized. For purposes of this Public License, the rights
	specified in Section 2(b)(1)-(2) are not Copyright and Similar
	Rights.
	d. Effective Technological Measures means those measures that, in the
	absence of proper authority, may not be circumvented under laws
	fulfilling obligations under Article 11 of the WIPO Copyright
	Treaty adopted on December 20, 1996, and/or similar international
	agreements.

	e. Exceptions and Limitations means fair use, fair dealing, and/or
	any other exception or limitation to Copyright and Similar Rights
	that applies to Your use of the Licensed Material.

	f. Licensed Material means the artistic or literary work, database,
	or other material to which the Licensor applied this Public
	License.

	g. Licensed Rights means the rights granted to You subject to the
	terms and conditions of this Public License, which are limited to
	all Copyright and Similar Rights that apply to Your use of the
	Licensed Material and that the Licensor has authority to license.

	h. Licensor means the individual(s) or entity(ies) granting rights
	under this Public License.

	i. NonCommercial means not primarily intended for or directed towards
	commercial advantage or monetary compensation. For purposes of
	this Public License, the exchange of the Licensed Material for
	other material subject to Copyright and Similar Rights by digital
	file-sharing or similar means is NonCommercial provided there is
	no payment of monetary compensation in connection with the
	exchange.

	j. Share means to provide material to the public by any means or
	process that requires permission under the Licensed Rights, such
	as reproduction, public display, public performance, distribution,
	dissemination, communication, or importation, and to make material
	available to the public including in ways that members of the
	public may access the material from a place and at a time
	individually chosen by them.

	k. Sui Generis Database Rights means rights other than copyright
	resulting from Directive 96/9/EC of the European Parliament and of
	the Council of 11 March 1996 on the legal protection of databases,
	as amended and/or succeeded, as well as other essentially
	equivalent rights anywhere in the world.

	l. You means the individual or entity exercising the Licensed Rights
	under this Public License. Your has a corresponding meaning.


	Section 2 -- Scope.

	a. License grant.

	1. Subject to the terms and conditions of this Public License,
	the Licensor hereby grants You a worldwide, royalty-free,
	non-sublicensable, non-exclusive, irrevocable license to
	exercise the Licensed Rights in the Licensed Material to:

	a. reproduce and Share the Licensed Material, in whole or
	in part, for NonCommercial purposes only; and

	b. produce, reproduce, and Share Adapted Material for
	NonCommercial purposes only.

	2. Exceptions and Limitations. For the avoidance of doubt, where
	Exceptions and Limitations apply to Your use, this Public
	License does not apply, and You do not need to comply with
	its terms and conditions.

	3. Term. The term of this Public License is specified in Section
	6(a).

	4. Media and formats; technical modifications allowed. The
	Licensor authorizes You to exercise the Licensed Rights in
	all media and formats whether now known or hereafter created,
	and to make technical modifications necessary to do so. The
	Licensor waives and/or agrees not to assert any right or
	authority to forbid You from making technical modifications
	necessary to exercise the Licensed Rights, including
	technical modifications necessary to circumvent Effective
	Technological Measures. For purposes of this Public License,
	simply making modifications authorized by this Section 2(a)
	(4) never produces Adapted Material.

	5. Downstream recipients.

	a. Offer from the Licensor -- Licensed Material. Every
	recipient of the Licensed Material automatically
	receives an offer from the Licensor to exercise the
	Licensed Rights under the terms and conditions of this
	Public License.

	b. No downstream restrictions. You may not offer or impose
	any additional or different terms or conditions on, or
	apply any Effective Technological Measures to, the
	Licensed Material if doing so restricts exercise of the
	Licensed Rights by any recipient of the Licensed
	Material.

	6. No endorsement. Nothing in this Public License constitutes or
	may be construed as permission to assert or imply that You
	are, or that Your use of the Licensed Material is, connected
	with, or sponsored, endorsed, or granted official status by,
	the Licensor or others designated to receive attribution as
	provided in Section 3(a)(1)(A)(i).

	b. Other rights.

	1. Moral rights, such as the right of integrity, are not
	licensed under this Public License, nor are publicity,
	privacy, and/or other similar personality rights; however, to
	the extent possible, the Licensor waives and/or agrees not to
	assert any such rights held by the Licensor to the limited
	extent necessary to allow You to exercise the Licensed
	Rights, but not otherwise.

	2. Patent and trademark rights are not licensed under this
	Public License.

	3. To the extent possible, the Licensor waives any right to
	collect royalties from You for the exercise of the Licensed
	Rights, whether directly or through a collecting society
	under any voluntary or waivable statutory or compulsory
	licensing scheme. In all other cases the Licensor expressly
	reserves any right to collect such royalties, including when
	the Licensed Material is used other than for NonCommercial
	purposes.


	Section 3 -- License Conditions.

	Your exercise of the Licensed Rights is expressly made subject to the
	following conditions.

	a. Attribution.

	1. If You Share the Licensed Material (including in modified
	form), You must:

	a. retain the following if it is supplied by the Licensor
	with the Licensed Material:

	i. identification of the creator(s) of the Licensed
	Material and any others designated to receive
	attribution, in any reasonable manner requested by
	the Licensor (including by pseudonym if
	designated);

	ii. a copyright notice;

	iii. a notice that refers to this Public License;

	iv. a notice that refers to the disclaimer of
	warranties;

	v. a URI or hyperlink to the Licensed Material to the
	extent reasonably practicable;

	b. indicate if You modified the Licensed Material and
	retain an indication of any previous modifications; and

	c. indicate the Licensed Material is licensed under this
	Public License, and include the text of, or the URI or
	hyperlink to, this Public License.

	2. You may satisfy the conditions in Section 3(a)(1) in any
	reasonable manner based on the medium, means, and context in
	which You Share the Licensed Material. For example, it may be
	reasonable to satisfy the conditions by providing a URI or
	hyperlink to a resource that includes the required
	information.

	3. If requested by the Licensor, You must remove any of the
	information required by Section 3(a)(1)(A) to the extent
	reasonably practicable.

	4. If You Share Adapted Material You produce, the Adapter's
	License You apply must not prevent recipients of the Adapted
	Material from complying with this Public License.


	Section 4 -- Sui Generis Database Rights.

	Where the Licensed Rights include Sui Generis Database Rights that
	apply to Your use of the Licensed Material:

	a. for the avoidance of doubt, Section 2(a)(1) grants You the right
	to extract, reuse, reproduce, and Share all or a substantial
	portion of the contents of the database for NonCommercial purposes
	only;

	b. if You include all or a substantial portion of the database
	contents in a database in which You have Sui Generis Database
	Rights, then the database in which You have Sui Generis Database
	Rights (but not its individual contents) is Adapted Material; and

	c. You must comply with the conditions in Section 3(a) if You Share
	all or a substantial portion of the contents of the database.

	For the avoidance of doubt, this Section 4 supplements and does not
	replace Your obligations under this Public License where the Licensed
	Rights include other Copyright and Similar Rights.


	Section 5 -- Disclaimer of Warranties and Limitation of Liability.

	a. UNLESS OTHERWISE SEPARATELY UNDERTAKEN BY THE LICENSOR, TO THE
	EXTENT POSSIBLE, THE LICENSOR OFFERS THE LICENSED MATERIAL AS-IS
	AND AS-AVAILABLE, AND MAKES NO REPRESENTATIONS OR WARRANTIES OF
	ANY KIND CONCERNING THE LICENSED MATERIAL, WHETHER EXPRESS,
	IMPLIED, STATUTORY, OR OTHER. THIS INCLUDES, WITHOUT LIMITATION,
	WARRANTIES OF TITLE, MERCHANTABILITY, FITNESS FOR A PARTICULAR
	PURPOSE, NON-INFRINGEMENT, ABSENCE OF LATENT OR OTHER DEFECTS,
	ACCURACY, OR THE PRESENCE OR ABSENCE OF ERRORS, WHETHER OR NOT
	KNOWN OR DISCOVERABLE. WHERE DISCLAIMERS OF WARRANTIES ARE NOT
	ALLOWED IN FULL OR IN PART, THIS DISCLAIMER MAY NOT APPLY TO YOU.

	b. TO THE EXTENT POSSIBLE, IN NO EVENT WILL THE LICENSOR BE LIABLE
	TO YOU ON ANY LEGAL THEORY (INCLUDING, WITHOUT LIMITATION,
	NEGLIGENCE) OR OTHERWISE FOR ANY DIRECT, SPECIAL, INDIRECT,
	INCIDENTAL, CONSEQUENTIAL, PUNITIVE, EXEMPLARY, OR OTHER LOSSES,
	COSTS, EXPENSES, OR DAMAGES ARISING OUT OF THIS PUBLIC LICENSE OR
	USE OF THE LICENSED MATERIAL, EVEN IF THE LICENSOR HAS BEEN
	ADVISED OF THE POSSIBILITY OF SUCH LOSSES, COSTS, EXPENSES, OR
	DAMAGES. WHERE A LIMITATION OF LIABILITY IS NOT ALLOWED IN FULL OR
	IN PART, THIS LIMITATION MAY NOT APPLY TO YOU.

	c. The disclaimer of warranties and limitation of liability provided
	above shall be interpreted in a manner that, to the extent
	possible, most closely approximates an absolute disclaimer and
	waiver of all liability.


	Section 6 -- Term and Termination.

	a. This Public License applies for the term of the Copyright and
	Similar Rights licensed here. However, if You fail to comply with
	this Public License, then Your rights under this Public License
	terminate automatically.

	b. Where Your right to use the Licensed Material has terminated under
	Section 6(a), it reinstates:

	1. automatically as of the date the violation is cured, provided
	it is cured within 30 days of Your discovery of the
	violation; or

	2. upon express reinstatement by the Licensor.

	For the avoidance of doubt, this Section 6(b) does not affect any
	right the Licensor may have to seek remedies for Your violations
	of this Public License.

	c. For the avoidance of doubt, the Licensor may also offer the
	Licensed Material under separate terms or conditions or stop
	distributing the Licensed Material at any time; however, doing so
	will not terminate this Public License.

	d. Sections 1, 5, 6, 7, and 8 survive termination of this Public
	License.


	Section 7 -- Other Terms and Conditions.

	a. The Licensor shall not be bound by any additional or different
	terms or conditions communicated by You unless expressly agreed.

	b. Any arrangements, understandings, or agreements regarding the
	Licensed Material not stated herein are separate from and
	independent of the terms and conditions of this Public License.


	Section 8 -- Interpretation.

	a. For the avoidance of doubt, this Public License does not, and
	shall not be interpreted to, reduce, limit, restrict, or impose
	conditions on any use of the Licensed Material that could lawfully
	be made without permission under this Public License.

	b. To the extent possible, if any provision of this Public License is
	deemed unenforceable, it shall be automatically reformed to the
	minimum extent necessary to make it enforceable. If the provision
	cannot be reformed, it shall be severed from this Public License
	without affecting the enforceability of the remaining terms and
	conditions.

	c. No term or condition of this Public License will be waived and no
	failure to comply consented to unless expressly agreed to by the
	Licensor.

	d. Nothing in this Public License constitutes or may be interpreted
	as a limitation upon, or waiver of, any privileges and immunities
	that apply to the Licensor or You, including from the legal
	processes of any jurisdiction or authority.

	=======================================================================

	Creative Commons is not a party to its public
	licenses. Notwithstanding, Creative Commons may elect to apply one of
	its public licenses to material it publishes and in those instances
	will be considered the “Licensor.” The text of the Creative Commons
	public licenses is dedicated to the public domain under the CC0 Public
	Domain Dedication. Except for the limited purpose of indicating that
	material is shared under a Creative Commons public license or as
	otherwise permitted by the Creative Commons policies published at
	creativecommons.org/policies, Creative Commons does not authorize the
	use of the trademark "Creative Commons" or any other trademark or logo
	of Creative Commons without its prior written consent including,
	without limitation, in connection with any unauthorized modifications
	to any of its public licenses or any other arrangements,
	understandings, or agreements concerning use of licensed material. For
	the avoidance of doubt, this paragraph does not form part of the
	public licenses.

	Creative Commons may be contacted at creativecommons.org.

	Project-URL: Homepage, https://github.com/facebookresearch/seamless_interaction
	Project-URL: Bug Tracker, https://github.com/facebookresearch/seamless_interaction
	Classifier: Programming Language :: Python :: 3
	Classifier: License :: OSI Approved :: CC BY-NC 4.0
	Classifier: Operating System :: OS Independent
	Requires-Python: >=3.10
	Description-Content-Type: text/markdown
	License-File: LICENSE
	Requires-Dist: boto3>=1.38.27
	Requires-Dist: datasets>=3.6.0
	Requires-Dist: fsspec>=2025.3.0
	Requires-Dist: jupyterlab>=4.4.3
	Requires-Dist: librosa>=0.11.0
	Requires-Dist: llvmlite>=0.43.0
	Requires-Dist: mypy>=1.16.1
	Requires-Dist: numpy>=1.26.0
	Requires-Dist: opencv-python>=4.7.0.72
	Requires-Dist: pandas>=1.5.1
	Requires-Dist: plotly>=6.1.2
	Requires-Dist: pre-commit>=4.2.0
	Requires-Dist: pylance>=0.29.0
	Requires-Dist: pytest>=8.4.1
	Requires-Dist: ruff>=0.12.0
	Requires-Dist: s3fs>=0.4.2
	Requires-Dist: streamlit>=1.46.0
	Requires-Dist: webdataset>=1.0.2
	Requires-Dist: wget>=3.2

	<div align="center">

	# Seamless Interaction Dataset

	<img src="./assets/banner.gif" alt="Seamless Interaction Dataset Banner" width="800px">

	A large-scale multimodal dataset of 4,000+ hours of human interactions for AI research


	<table>
	<tr>
	<td align="center">
	<a href="https://ai.meta.com/blog/seamless-interaction-natural-conversational-dynamics/">
	🖼️ Blog
	</a>
	</td>
	<td align="center">
	<a href="https://ai.meta.com/research/seamless-interaction/">
	🌐 Website
	</a>
	</td>
	<td align="center">
	<a href="https://www.aidemos.meta.com/seamless_interaction_dataset">
	🎮 Demo
	</a>
	</td>
	<td align="center">
	<a href="https://huggingface.co/datasets/facebook/seamless-interaction">
	🤗 HuggingFace
	</a>
	</td>
	<td align="center">
	<a href="https://ai.meta.com/research/publications/seamless-interaction-dyadic-audiovisual-motion-modeling-and-large-scale-dataset">
	📄 Paper
	</a>
	</td>
	</tr>
	</table>


	</div>

	## Overview

	Human communication involves a complex interplay of verbal and nonverbal signals, essential for conveying meaning and achieving interpersonal goals.

	The Seamless Interaction Dataset is a large-scale collection of over 4,000 hours of face-to-face interaction footage from more than 4,000 participants in diverse contexts.
	This dataset enables the development of AI technologies that understand human interactions and communication, unlocking breakthroughs in:

	- 🤖 Virtual agents and embodied AI
	- 🎭 Natural human-computer interaction
	- 📡 Advanced telepresence experiences
	- 📊 Multimodal content analysis tools
	- 🎬 Animation and synthetic content generation

	## 🚀 Quick Start

	```bash
	git clone https://github.com/facebookresearch/seamless_interaction
	cd seamless_interaction
	pip install -e .
	streamlit run src/seamless_interaction/app/Welcome.py

	# if you use uv
	uv sync
	uv run streamlit run src/seamless_interaction/app/Welcome.py
	```

	Explore the dataset with our interactive browser:

	Features:
	- 🔍 Hierarchical Navigation: Browse by Label → Split → Batch → Interaction
	- 🎲 Random Sampling: Discover interactions with one-click random selection
	- 📥 Download Interface: Download specific batches with size estimation and progress tracking
	- 🎬 Video Viewer: Side-by-side participant videos with synchronized playback
	- 📊 Data Analysis: Overview statistics and distribution plots
	- 📁 File Management: Organize and preview audio, JSON, and NPZ files with expandable dropdowns

	### Download Options

	We provide comprehensive download methods supporting all research scales and requirements:

	\| Scale \| Size \| Method \| Use Case \| Script \| Sampling \|
	\|-----------\|----------\|------------\|--------------\|------------\|-------------\|
	\| 🔍 Single Example \| ~100MB \| S3 \| Quick exploration, understanding data structure \| [`download_s3.py`](./scripts/download_s3.py#L10) \| Auto-sample from preferred vendors \|
	\| 👥 Interaction Pair \| ~200MB \| S3 \| Study conversational dynamics between participants \| [`download_s3.py`](./scripts/download_s3.py#L34) \| Auto-detect conversation pairs \|
	\| 📂 Sample Set \| ~1GB \| S3/HF \| Initial prototyping, algorithm development \| [`download_s3.py`](./scripts/download_s3.py#L66), [`download_hf.py`](./scripts/download_hf.py#L10) \| File selection or archive-based \|
	\| 🎯 Session Groups \| ~400MB \| S3 \| Deep conversational context, session dynamics \| [`download_s3.py`](./scripts/download_s3.py#L100) \| Auto-sample rich sessions \|
	\| 📦 Single Batch \| ~50GB \| HF \| Substantial local development, full exploration \| [`download_hf.py`](./scripts/download_hf.py#L24) \| WebDataset tarball download \|
	\| 🗂️ Multiple Batches \| ~150GB+ \| HF \| Training datasets, large-scale analysis \| [`download_hf.py`](./scripts/download_hf.py#L38) \| WebDataset tarball download \|
	\| 🎯 Different Splits \| Variable \| HF \| Cross-validation (train/dev/test, improvised/naturalistic) \| [`download_hf.py`](./scripts/download_hf.py#L55) \| WebDataset tarball download \|
	\| 🌍 Whole Dataset \| ~27TB \| HF \| Complete research dataset, production systems \| [`download_hf.py`](./scripts/download_hf.py#L82) \| WebDataset tarball download \|

	#### 🔍 S3 Download - Fine-grained exploration
	Perfect for exploring individual interactions or specific file IDs. Downloads from S3 and automatically converts to consistent format (.wav, .mp4, .npz, .json).

	```python
	# Initialize with configuration for cleaner setup
	from seamless_interaction.fs import SeamlessInteractionFS, DatasetConfig

	config = DatasetConfig(
	label="improvised",
	split="dev",
	preferred_vendors_only=True,
	local_dir=Path.home() / "datasets/seamless_interaction", # note: we will automatically create the directory if it doesn't exist
	)
	fs = SeamlessInteractionFS(config=config)
	# Or use defaults: fs = SeamlessInteractionFS()

	file_ids = fs.sample_random_file_ids(num_samples=1)
	fs.gather_file_id_data_from_s3(file_ids[0])
	# Or specify exact file: fs.gather_file_id_data_from_s3("V00_S0809_I00000582_P0947")

	# Files are organized as:
	# local_dir/improvised/train/0000/0005/V00_S0809_I00000582_P0947.wav
	# local_dir/improvised/train/0000/0005/V00_S0809_I00000582_P0947.mp4
	# local_dir/improvised/train/0000/0005/V00_S0809_I00000582_P0947.json
	# local_dir/improvised/train/0000/0005/V00_S0809_I00000582_P0947.npz
	```

	For more details, please refer to the [S3 Download Example](./scripts/download_s3.py).

	#### 📦 HuggingFace Download - Batch exploration
	Ideal for downloading self-contained batches (~50GB each) for local exploration. Each batch contains complete interaction pairs.

	```python
	from seamless_interaction.fs import SeamlessInteractionFS, DatasetConfig

	# Initialize with configuration
	config = DatasetConfig(label="improvised", split="dev")
	fs = SeamlessInteractionFS(config=config)

	# Sample set (~1GB) - Quick exploration on laptops
	fs.download_batch_from_hf(batch_idx=0, archive_list=[0])

	# Single batch (~50-100GB) - Substantial local development
	fs.download_batch_from_hf(batch_idx=0)
	```

	For more details, please refer to the [HuggingFace Download Example](./scripts/download_hf.py).

	### Working with Downloaded Data

	```python
	from seamless_interaction.fs import SeamlessInteractionFS
	import json
	import numpy as np
	import cv2
	import librosa

	# Load interaction data
	def load_interaction_data(file_id):
	"""Load all modalities for a given file ID."""

	fs = SeamlessInteractionFS()
	paths = fs.get_path_list_for_file_id_local(file_id)
	print(paths)

	data = {}
	for path in paths:
	if path.endswith('.mp4'):
	data['video'] = cv2.VideoCapture(path)
	elif path.endswith('.wav'):
	data['audio'], data['sample_rate'] = librosa.load(path, sr=48_000)
	elif path.endswith('.json'):
	with open(path) as f:
	data['json'] = json.load(f)
	elif path.endswith('.npz'):
	data['npz'] = np.load(path)

	return data

	fs.download_archive_from_hf(
	archive=0,
	label="improvised",
	split="test",
	batch=0,
	local_dir=None,
	extract=True,
	)
	# Example usage
	interaction = load_interaction_data("V01_S0223_I00000127_P1505")
	print(f"Available feature keys: {list(interaction['npz'].keys())}")
	print(f"Right hand pose data shape: {interaction['npz']['smplh:right_hand_pose'].shape}")
	```

	### Basic Data Loading (HF + WebDataset)

	```python
	from datasets import load_dataset

	# configure
	label = "improvised"
	split = "dev"
	batch_idx = 0
	archive_list = [0, 1]

	base_url = (
	f"https://huggingface.co/datasets/facebook/"
	f"seamless-interaction/resolve/main/{label}/{split}/"
	"{batch_idx:04d}/{archive_idx:04d}.tar"
	)
	urls = [base_url.format(batch_idx=batch_idx, archive_idx=archive_idx) for archive_idx in archive_list]
	dataset = load_dataset(
	"webdataset", data_files={split: urls}, split=split, streaming=True
	)

	for item in dataset:
	break

	isinstance(item["mp4"], bytes)
	# True
	item["npz"].keys()
	# dict_keys(['boxes_and_keypoints:box', 'boxes_and_keypoints:is_valid_box', 'boxes_and_keypoints:keypoints', 'movement:EmotionArousalToken', 'movement:EmotionValenceToken', 'movement:FAUToken', 'movement:FAUValue', 'movement:alignment_head_rotation', 'movement:alignment_translation', 'movement:emotion_arousal', 'movement:emotion_scores', 'movement:emotion_valence', 'movement:expression', 'movement:frame_latent', 'movement:gaze_encodings', 'movement:head_encodings', 'movement:hypernet_features', 'movement:is_valid', 'smplh:body_pose', 'smplh:global_orient', 'smplh:is_valid', 'smplh:left_hand_pose', 'smplh:right_hand_pose', 'smplh:translation'])
	item["json"].keys()
	# dict_keys(['id', 'metadata:transcript', 'metadata:vad'])
	item["wav"].keys()
	# dict_keys(['path', 'array', 'sampling_rate'])
	```

	Check out the [dataloader_webdataset.py](./scripts/dataloader_webdataset.py) script for more details.


	## 🔍 Description

	The `seamless_interaction` repository is split into several main components:

	### 📊 Dataset

	The repository provides comprehensive tools for downloading, processing, and utilizing the Seamless Interaction dataset for research and development. The dataset includes:

	- Raw and processed multimodal data: Video, audio, transcripts, and annotations
	- Precomputed features: Motion capture, facial keypoints, voice activity detection
	- Metadata: Participant personality (BFI-2), interaction contexts, and relationships

	### 📂 Repository Structure

	```
	seamless_interaction/
	├── assets/ # Static assets for documentation
	│ ├── banner.png
	│ └── filelist.csv # File list for the dataset
	├── scripts/ # Example scripts for dataset usage
	│ ├── dataloader_webdataset.py
	│ ├── download_hf.py
	│ ├── download_s3.py
	├── src/seamless_interaction/ # Main package source code
	│ ├── app/ # Data exploration application
	│ ├── fs.py # Filesystem interface for dataset access
	│ ├── utils.py # General utility functions
	│ ├── constants.py # Dataset constants and configuration
	│ └── __init__.py
	├── LICENSE # CC-BY-NC 4.0 license
	└── pyproject.toml # Python package configuration
	```

	## 📦 Deep Dive into the Dataset

	### Dataset Structure

	The Seamless Interaction Dataset is organized into two main categories/labels:
	- Improvised: Interactions primarily based on predefined scenarios with guided prompts with at least one professional actor.
	- Naturalistic: Prompted conversations that can be carried out by normal people.

	```
	seamless_interaction
	├── improvised # Interactions with guided prompts
	│ ├── dev
	│ │ ├── 1P-IS/ # First-party internal state annotations
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.json
	│ │ ├── 1P-R/ # First-party internal state rationale annotations
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.json
	│ │ ├── 3P-IS/ # Third-party internal state annotations
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.json
	│ │ ├── 3P-R/ # Third-party internal state rationale annotations
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.json
	│ │ ├── 3P-V/ # Third-party visual annotation
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.json
	│ │ ├── audio/ # Speaker-bleed denoised audio
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.wav
	│ │ ├── boxes_and_keypoints/
	│ │ │ ├── box/ # Bounding boxes for each participant
	│ │ │ ├── is_valid_box/ # Whether bounding boxes are valid
	│ │ │ └── keypoints/ # Detected facial/body keypoints
	│ │ ├── movement/ # Quantified Imitator movement features
	│ │ │ ├── emotion_arousal/ # Arousal measures
	│ │ │ ├── emotion_valence/ # Valence measures
	│ │ │ ├── emotion_scores/ # Emotion detection scores
	│ │ │ ├── expression/ # Facial expression parameters
	│ │ │ ├── FAUToken/ # Facial Action Unit tokens
	│ │ │ ├── FAUValue/ # Facial Action Unit values
	│ │ │ ├── gaze_encodings/ # Eye gaze direction encodings
	│ │ │ ├── head_encodings/ # Head position/rotation encodings
	│ │ │ ├── frame_latent/ # Per-frame latent representations
	│ │ │ └── is_valid/ # Validity flags for extracted features
	│ │ ├── smplh/ # SMPL-H body model parameters
	│ │ │ ├── body-pose/ # Body pose parameters
	│ │ │ ├── global_orient/ # Global orientation parameters
	│ │ │ ├── is_valid/ # Valid frames indicators
	│ │ │ ├── left_hand_pose/ # Left hand pose parameters
	│ │ │ ├── right_hand_pose/ # Right hand pose parameters
	│ │ │ └── translation/ # Global translation parameters
	│ │ ├── transcript/ # Time-aligned speech transcription
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.jsonl
	│ │ ├── vad/ # Voice activity detection
	│ │ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.jsonl
	│ │ └── video/ # Raw HD video recordings
	│ │ └── V<vendor>_S<session>_I<interaction>_P<participant>.mp4
	│ ├── test/ # Test split with similar structure
	│ └── train/ # Training split with similar structure
	└── naturalistic/ # Spontaneous conversations
	├── dev/ # Same structure as improvised/dev
	├── test/ # Same structure as improvised/test
	└── train/ # Same structure as improvised/train
	```

	Each file is named according to a consistent convention:
	- `V<vendor_id>`: Collection site/vendor identifier
	- `S<session_id>`: Unique session identifier
	- `I<interaction_id>`: Specific interaction within a session
	- `P<participant_id>`: Individual participant identifier

	### Available Modalities and Features

	Each interaction in the dataset includes:

	\| Modality \| Description \| File Format \| Sample Rate \|
	\|----------\|-------------\|-------------\|-------------\|
	\| 🎥 Video \| High-definition face-to-face footage \| MP4 (H.264) \| 30/29.97 FPS, 1080p \|
	\| 🎙️ Audio \| Denoised audio with separate channels \| WAV \| 48kHz, 16-bit \|
	\| 📝 Transcript \| Time-aligned speech transcription \| JSONL \| - \|
	\| 🏃 SMPL-H \| 3D body model parameters \| NPY \| 30 Hz \|
	\| 🧠 Imitator Movement Features \| Comprehensive quantified imitator movement data \| NPY \| 30 Hz \|
	\| 📊 Annotations \| Human-annotated behavioral data \| JSON \| - \|
	\| 🔊 VAD \| Voice activity detection \| JSONL \| 100 Hz \|
	\| 📦 Keypoints \| Face and body keypoints \| NPY \| 30 Hz \|

	#### Annotation Types

	The dataset includes several types of human annotations for rich behavioral analysis:

	\| Annotation \| Hours \| Total Annotations \| Mean # Tokens \|
	\|------------\|-------------\|--------\|--------\|
	\| 1P-IS (1st-party internal state annotations) \| 1.1 \| 751 \| 5.8 \|
	\| 1P-R (1st-party internal state rationale annotations) \| 1.1 \| 751 \| 10.2 \|
	\| 3P-IS (3rd-party internal state annotations) \| 4.7 \| 5132 \| 5.2 \|
	\| 3P-R (3rd-party internal state rationale annotations) \| 4.7 \| 5132 \| 11.3 \|
	\| 3P-V (3rd-party visual annotation) \| 4.7 \| 5132 \| 14.6 \|

	Please refer to the [technical report](https://ai.meta.com/research/publications/seamless-interaction-dyadic-audiovisual-motion-modeling-and-large-scale-dataset/) for a more detailed overview of annotations.

	#### Movement/Imitator Feature Types

	The movement directory contains rich behavioral features (output of the Imitator model):

	\| Feature \| Description \|
	\|---------\|-------------\|
	\| `emotion_arousal` \| Arousal intensity measurements \|
	\| `emotion_valence` \| Valence (positive/negative) measurements \|
	\| `emotion_scores` \| Detected emotion categorical scores \|
	\| `expression` \| Parametric facial expression encodings \|
	\| `FAUToken`/`FAUValue` \| Facial Action Unit tokens and intensity values \|
	\| `gaze_encodings` \| Neural encodings of gaze direction \|
	\| `head_encodings` \| Neural encodings of head position and rotation \|
	\| `frame_latent` \| Per-frame latent representations \|
	\| `alignment_head_rotation` \| Head rotation data for temporal alignment \|
	\| `alignment_translation` \| Translation parameters for temporal alignment \|
	\| `EmotionArousalToken`/`EmotionValenceToken` \| Discretized emotion tokens \|
	\| `hypernet_features` \| Features from hypernetwork processing \|

	### Download Strategy Guide

	We provide two complementary download methods optimized for different research workflows:

	\| Method \| Use Case \| Best For \| Download Size \| Parallelization \|
	\|--------\|----------\|----------\|---------------\|-----------------\|
	\| S3 Direct \| Fine-grained exploration \| Individual interactions, interaction pairs \| Per file (~100MB) \| ✅ Multiprocessing \|
	\| HuggingFace Batches \| Batch processing \| Local dataset exploration, model training \| ~50-100GB per batch \| ✅ Multiprocessing \|

	#### When to Use S3 Download
	- Qualitative analysis: Examining specific interactions in detail
	- Pair studies: Analyzing conversational dynamics between participants
	- Feature exploration: Understanding data structure before large downloads
	- Development: Testing code with minimal data

	#### When to Use HuggingFace Download
	- Model training: Need substantial training data
	- Batch processing: Analyzing patterns across many interactions
	- Local exploration: Want self-contained dataset on laptop/workstation
	- Reproducible research: Ensure consistent data splits

	#### Performance Optimization

	```python
	# Optimal settings for different systems
	config_default = DatasetConfig() # auto-detects system resources
	config_laptop = DatasetConfig(num_workers=4) # Laptop/small workstation
	config_workstation = DatasetConfig(num_workers=8) # High-end workstation
	config_server = DatasetConfig(num_workers=16) # Server/cluster node

	# Memory-efficient batch processing
	config = DatasetConfig(label="improvised", split="train")
	fs = SeamlessInteractionFS(config=config)
	for batch_idx in range(10): # Process in chunks
	fs.download_batch_from_hf(batch_idx=batch_idx)
	# Process batch here...
	# Delete batch to free space if needed
	```

	### Dataset Versions

	The dataset is organized in self-contained batches for flexible exploration:

	\| Split \| Batches \| Size per Batch \| Total Size \| Description \|
	\|-------\|---------\|----------------\|------------\|-------------\|
	\| dev \| 5 \| ~50GB \| ~500GB \| Development/validation set \|
	\| test \| 5 \| ~50GB \| ~500TB \| Hold-out test set \|
	\| train \| 200+ \| ~50GB \| ~20TB+ \| Full training data \|

	#### Recommended Download Strategies

	```python
	# Strategy 1: Quick Start (Laptop-friendly)
	config = DatasetConfig(label="improvised", split="dev")
	fs = SeamlessInteractionFS(config=config)
	fs.download_batch_from_hf(batch_idx=0, archive_list=[0, 1, 2]) # ~6GB

	# Strategy 2: Research Dataset (Workstation)
	config = DatasetConfig(label="improvised", split="dev")
	fs = SeamlessInteractionFS(config=config)
	fs.download_batch_from_hf(batch_idx=0) # Full dev set ~50-100GB

	config = DatasetConfig(label="naturalistic", split="dev")
	fs = SeamlessInteractionFS(config=config)
	fs.download_batch_from_hf(batch_idx=0) # Both interaction types

	# Strategy 3: Production Training (Server/Cluster)
	config = DatasetConfig(label="improvised", split="train")
	fs = SeamlessInteractionFS(config=config)
	for batch_idx in range(20): # First 20 training batches (~1TB)
	fs.download_batch_from_hf(batch_idx=batch_idx)
	```

	#### File Format Specifications

	Our data is stored in the following formats for optimal usability:

	\| Format \| Description \| Usage \|
	\|--------\|-------------\|-------\|
	\| NPZ \| NumPy array files \| Efficient storage of numerical feature vectors, keypoints, and parameters \|
	\| JSONL \| JSON Lines \| Time-aligned annotations with one event per line (e.g., transcripts, VAD) \|
	\| JSON \| JavaScript Object Notation \| Structured metadata and annotations with timestamps \|
	\| MP4 \| MPEG-4 Part 14 \| High-quality compressed video with H.264 encoding \|
	\| WAV \| Waveform Audio \| Uncompressed audio for highest fidelity processing \|

	## 🧪 Research Applications

	The Seamless Interaction Dataset enables research across multiple domains:

	### Embodied AI and Virtual Agents
	- Train agents that display natural gestures
	- Model turn-taking dynamics and interaction rhythms
	- Generate contextually appropriate responses to human behavior

	### Multimodal Understanding
	- Analyze cross-modal correlations between speech, gesture, and expressions
	- Extract behavioral patterns from large-scale interaction data
	- Develop models to understand social dynamics

	### Human-Computer Interaction
	- Design interfaces that respond to subtle human cues
	- Improve telepresence technologies with better behavioral modeling
	- Create more natural conversational agents

	### Animation and Content Creation
	- Generate realistic human behaviors for animated characters
	- Synthesize conversational dynamics for virtual production
	- Create training data for digital human technologies

	## ⚠️ Known Limitations and Noise in Metadata

	Given the scale and complexity involved in collecting the Seamless Interaction dataset, there are several known limitations that we will address in our ongoing work, with improvements planned for in future versions:

	### Errors in Human-Based Time-Stamping
	The core unit of the dataset is interactions. An interaction defines the active time during which a participant’s conversation and behavior can be linked to a pair of prompts. We have observed instances of misaligned time-stamps, including:
	- Annotated start/end times may be too early or too late.
	- Occasional misalignment between prompt text and spoken material.
	- Ordering of prompts that may contain off-by-one errors.

	Despite our efforts to automatically identify and correct these errors, approximately 10% of the interactions remain affected.

	### Time Stamping "Noise" in Moments of Interest (MOI)
	While defining a MOI inherently involves some subjectivity, there are rare instances where:
	- The described behavior only represents a subset of the observed behavior.
	- The duration of the MOI does not fully capture the annotated behavior.

	### Incorrect Assignment of Participant IDs
	In rare instances, we have observed:
	- Duplicate participant identifiers being assigned to different individuals.
	- The same individual being mapped to different identifiers.

	### Unreleased "Meta Time"
	Currently, the dataset only contains active time segments - time in which two participants are actively responding to prompts. Meta time refers to the time between active segments in which participants are studying their new prompts, taking a break, etc. Meta time constitutes hundreds of hours in the raw collection and maybe be explored for future releases.

	### Variation in Recording Site Consistency
	This multi-site project contains variation in:
	- Recording quality, including issues like speaker bleed and participants staying in frame.
	- Acting quality in Improvised segments.
	- The likelihood of time-stamping errors.

	All vendors met our technical requirements; however,there is noticeable variation in production quality across different sites.

	## 🤝 Contributing

	We welcome contributions from the research community! Here are some ways to contribute:

	- Bug Reports & Feature Requests: Open issues on GitHub
	- Dataset Improvements: Help enhance our preprocessing pipelines or annotations
	- Model Contributions: Submit your models to our benchmarks
	- Documentation: Improve our guides, tutorials, and API documentation
	- Sample Code: Share example applications built with the dataset

	Please see our [CONTRIBUTING.md](CONTRIBUTING.md) for detailed guidelines, code of conduct, and submission processes.

	## 📄 License & Data Usage Policy

	The Seamless Interaction Dataset is licensed under CC-BY-NC 4.0 (Creative Commons Attribution-NonCommercial 4.0 International).

	This means you are free to:
	- Share — copy and redistribute the material in any medium or format
	- Adapt — remix, transform, and build upon the material

	Under the following terms:
	- Attribution — You must give appropriate credit, provide a link to the license, and indicate if changes were made.
	- NonCommercial — You may not use the material for commercial purposes without explicit permission.


	## 📑 Citation

	If you use the Seamless Interaction Dataset in your research, please cite:

	<details>
	<summary>BibTeX</summary>

	```bibtex
	@article{seamless_interaction,
	title={Seamless Interaction: Dyadic Audiovisual Motion Modeling and Large-Scale Dataset},
	author={Vasu Agrawal and
	Akinniyi Akinyemi and
	Kathryn Alvero and
	Morteza Behrooz and
	Julia Buffalini and
	Fabio Maria Carlucci and
	Joy Chen and
	Junming Chen and
	Zhang Chen and
	Shiyang Cheng and
	Praveen Chowdary and
	Joe Chuang and
	Antony D'Avirro and
	Jon Daly and
	Ning Dong and
	Mark Duppenthaler and
	Cynthia Gao and
	Jeff Girard and
	Martin Gleize and
	Sahir Gomez and
	Hongyu Gong and
	Srivathsan Govindarajan and
	Brandon Han and
	Sen He and
	Denise Hernandez and
	Yordan Hristov and
	Rongjie Huang and
	Hirofumi Inaguma and
	Somya Jain and
	Raj Janardhan and
	Qingyao Jia and
	Christopher Klaiber and
	Dejan Kovachev and
	Moneish Kumar and
	Hang Li and
	Yilei Li and
	Pavel Litvin and
	Wei Liu and
	Guangyao Ma and
	Jing Ma and
	Martin Ma and
	Xutai Ma and
	Lucas Mantovani and
	Sagar Miglani and
	Sreyas Mohan and
	Louis-Philippe Morency and
	Evonne Ng and
	Kam-Woh Ng and
	Tu Anh Nguyen and
	Amia Oberai and
	Benjamin Peloquin and
	Juan Pino and
	Jovan Popovic and
	Omid Poursaeed and
	Fabian Prada and
	Alice Rakotoarison and
	Alexander Richard and
	Christophe Ropers and
	Safiyyah Saleem and
	Vasu Sharma and
	Alex Shcherbyna and
	Jia Shen and
	Jie Shen and
	Anastasis Stathopoulos and
	Anna Sun and
	Paden Tomasello and
	Tuan Tran and
	Arina Turkatenko and
	Bo Wan and
	Chao Wang and
	Jeff Wang and
	Mary Williamson and
	Carleigh Wood and
	Tao Xiang and
	Yilin Yang and
	Zhiyuan Yao and
	Chen Zhang and
	Jiemin Zhang and
	Xinyue Zhang and
	Jason Zheng and
	Pavlo Zhyzheria and
	Jan Zikes and
	Michael Zollhoefer
	},
	url={https://ai.meta.com/research/publications/seamless-interaction-dyadic-audiovisual-motion-modeling-and-large-scale-dataset/},
	year={2025}
	}
	```
	</details>

	## 🙏 Acknowledgments

	This project was made possible thanks to contributions from:

	- The thousands of participants who provided interaction data
	- Our dedicated annotation and QA team
	- Research collaborators from multiple institutions
	- FAIR (Fundamental AI Research)
	- The open-source community for valuable tools and libraries
	- Our data collection partners across multiple sites
	- Meta Reality Labs for supporting this research initiative