Dataset Card for RetailAction

RetailAction is a large-scale dataset for multi-view spatio-temporal localization of human-object interactions in real-world retail environments. It contains 21,000 dual-camera samples (42,000 synchronized top-view video clips, ~41 hours total) captured across 10 U.S. convenience stores. Each sample carries point-based annotations marking the precise location where a customer touches a product, along with temporal boundaries and action class (take/put/touch). The FiftyOne dataset is a grouped video dataset with two slices per group (rank0, rank1) and includes frame-level 24-joint body pose keypoints and face positions for the labeled subject.

Installation

pip install -U fiftyone

Usage

import fiftyone as fo
from fiftyone.utils.huggingface import load_from_hub

# Load the dataset
dataset = load_from_hub("harpreetsahota/RetailAction")

# Launch the App
session = fo.launch_app(dataset)

Dataset Details

Dataset Description

RetailAction addresses a gap in retail-specific multi-view spatio-temporal localization. Unlike prior action recognition datasets that either lack spatio-temporal annotations or are captured in controlled lab settings, RetailAction provides large-scale, real-world recordings from operational convenience stores. Instead of annotating person bounding boxes, it marks the precise 2D point where a customer's hand contacts a product — an interaction-centric paradigm suited to retail analytics and autonomous checkout applications.

The dataset was collected using 360-degree ceiling-mounted cameras in 10 stores across the United States. An automated pipeline based on 3D pose tracking and a kinematic Graph Convolutional Network identified candidate interaction intervals. The two best camera views were selected per interaction, and frames were subsampled to at most 32 using a velocity-and-acceleration-based scoring model to prioritize moments of significant hand movement.

• Curated by: Standard AI (standard-cognition)
• Funded by: Standard AI
• Shared by: Standard AI
• Language(s): English
• License: Standard AI custom license (see LICENSE file in the source repository)

Dataset Sources

• Repository: https://huggingface.co/datasets/standard-cognition/RetailAction
• Paper: RetailAction: Dataset for Multi-View Spatio-Temporal Localization of Human-Object Interactions in Retail — ICCV 2025 Retail Vision Workshop
• Demo: [More Information Needed]

Uses

Direct Use

• Training and evaluating multi-view spatio-temporal action localization models
• Benchmarking spatial and temporal localization accuracy in retail environments
• Shopper behavior analysis and customer-product interaction research
• Development of autonomous checkout systems
• Research into human-object interaction understanding from top-view cameras

Out-of-Scope Use

• Re-identification of individuals in the videos
• Any use that conflicts with the Standard AI custom license included in the source repository
• Applications outside the retail and computer vision research domains for which the dataset was designed

Dataset Structure

FiftyOne Dataset Topology

This is a grouped video dataset. Each group represents one retail interaction event captured by two ceiling-mounted cameras simultaneously. The two slices are named rank0 (default) and rank1, corresponding to the highest-ranked and second-highest-ranked camera views selected by the collection pipeline.

Property	Value
Dataset type	Grouped video
Groups	21,000
Slices per group	2 (rank0, rank1)
Total samples	42,000
Media type	Video (H.264, 600×600 px, 6 fps, ≤32 frames, ≤5.33 s per clip)
Default slice	rank0
Splits	train (17,222), validation (1,277), test (2,501)

Splits are defined by unique shopper identity: all clips of a given individual appear exclusively in one split to prevent data leakage.

Sample-Level Fields

These fields are identical across both slices of every group.

Field	FiftyOne type	Description
split	StringField	Dataset split: "train", "validation", or "test"
sample_id	StringField	Zero-padded source folder ID (e.g. "000042")
segment_duration_s	FloatField	Clip duration in seconds (~2.7–10.2 s, variable across splits)
n_actions	IntField	Number of annotated interactions in the clip (0–4+)
action_classes	ListField(StringField)	Sorted list of unique action labels present in the clip
ground_truth	TemporalDetections	Labeled interactions with temporal support and spatial interaction point

Per-Slice Frame-Level Fields

These fields are attached independently to each video slice since the two cameras see different perspectives.

Field	FiftyOne type	Description
pose	Keypoints	24-joint body pose of the labeled subject (one Keypoint per frame)
face_position	Keypoints	Face center of the labeled subject (one Keypoint per frame)

Label Types and Rationale

ground_truth → fo.TemporalDetections

Each annotated interaction becomes one fo.TemporalDetection with:

• label: action class — "take", "put", or "touch"
• support: [first_frame, last_frame] (1-indexed frame numbers, derived by mapping the normalized [0, 1] temporal interval from the metadata to actual frame timestamps)
• .x, .y: normalized [0, 1] 2D interaction point for this camera view (stored as dynamic attributes on the detection so they travel with the label)

TemporalDetections was chosen over per-frame Detections because the annotations describe temporal segments, not individual frames. The spatial point is stored as an attribute rather than a separate Keypoints field so the spatial and temporal labels remain co-located on the same object.

pose → fo.Keypoints (per frame)

Each frame with a detected subject carries one fo.Keypoint with 24 points (one per joint), ordered to match dataset.skeletons["pose"]. A fo.KeypointSkeleton is registered on the dataset with the full joint name list and 10 bone chains for App visualization.

The source metadata stores joint coordinates as [row, col] (y, x). The ingest script swaps them to FiftyOne's (x, y) convention. Joints absent from the pose dict are stored as (nan, nan) to preserve index alignment without drawing phantom skeleton edges. Per-joint confidence scores are stored in Keypoint.confidence and can be used to filter detections at query time:

from fiftyone import ViewField as F
view = dataset.filter_keypoints("pose", F("confidence") > 0.3)

Note on joint count: The source README and paper claim 18 body keypoints. The actual data ships 24: top_of_head, nose, neck, {right,left}_{shoulder,elbow,wrist,hand,hip,knee,ankle,foot}, middle_of_waist, {right,left}_{eye,ear}.

face_position → fo.Keypoints (per frame)

The face center (col, row) from the metadata maps directly to FiftyOne's (x, y) — no coordinate swap needed. This field covers only the labeled subject; other people in frame have no face data attached.

Pose Skeleton

dataset.skeletons["pose"]  # fo.KeypointSkeleton with 24 labels and 10 bone chains

Bone chains:

• top_of_head → nose → neck → middle_of_waist
• nose → right_eye → right_ear
• nose → left_eye → left_ear
• neck → right_shoulder → right_elbow → right_wrist → right_hand
• neck → left_shoulder → left_elbow → left_wrist → left_hand
• right_shoulder → right_hip
• left_shoulder → left_hip
• right_hip → middle_of_waist → left_hip
• right_hip → right_knee → right_ankle → right_foot
• left_hip → left_knee → left_ankle → left_foot

Parsing Decisions and Known Data Quality Issues

The following issues were discovered by inspecting the actual files (not documented in the source README):

Issue	Scope	Handling
Individual pose entries (poses[i]) are None	All splits, ~10% of frames	Skipped — no pose label attached to that frame
frame_timestamps list is None	~11.8% of train cam entries	Synthetic evenly-spaced timestamps generated from segment_duration_s
poses key absent entirely from cam entry	148/34,444 train cam entries	Frame labels skipped for that slice; warned in log
README claims 18 joints; data has 24	All splits	All 24 joints stored; KeypointSkeleton reflects actual count

Action Distribution

Action	Frequency	Count (approx.)
take	97.2%	~20,100
put	~2%	~420
touch	<1%	~210

The heavy take imbalance reflects real customer behavior; the distribution was not artificially adjusted. Training models on this data requires weighted sampling or loss weighting for put and touch.

Evaluation

The competition task (PRAW @ WACV 2027, Track 3: TV-STAL) evaluates predictions with a COCO-style mAP adapted for 2D point localization:

• mAP: averaged over spatial distance and temporal IoU thresholds
• mAPs: spatial Average Precision (averaged over lowest temporal IoU threshold)
• mAPt: temporal Average Precision (averaged over highest spatial distance threshold)
• Spatial matching uses a per-video pixels-to-meters factor derived from pose bone lengths

Published baselines (test set):

Model	Type	mAP	mAPs	mAPt
MoViNet-A2	Conv	33.5	43.8	60.9
SlowFast-R101	Conv	40.2	50.4	53.2
MViT-b	Transf	41.7	55.6	58.2
ViT-small	Transf	28.3	42.4	46.9
ViT-base	Transf	31.1	45.7	47.0
ViT-giant (frozen)	Transf	38.5	50.3	58.0

Dataset Creation

Curation Rationale

RetailAction was created to bridge the gap between generic action recognition benchmarks and the specific demands of retail AI. Existing datasets lack the combination of real-world store recordings, multi-view coverage, and fine-grained interaction-point annotations needed for applications like autonomous checkout and shopper behavior analysis. The dataset introduces an interaction-centric annotation paradigm — marking where a hand contacts a product rather than drawing a person bounding box — which is more informative for downstream retail applications.

Source Data

Data Collection and Processing

Data was collected in 10 medium-to-small convenience stores across the United States over multiple years, capturing actual (unscripted) customer visits. Recording devices were 360-degree top-view cameras operating at 30 FPS with a 2880×2880 resolution, ceiling-mounted at an average height of 2.5 meters and positioned so every point in the area of interest is covered by at least two cameras.

An automated pipeline processed the raw footage:

1. 2D pose estimation using a proprietary version of PersonLab [Papandreou et al., ECCV 2018] fine-tuned on 360-degree store footage.
2. Multi-view 3D pose reconstruction via tracklet clustering and triangulation.
3. Interaction interval detection using a kinematic Graph Convolutional Network [Yan et al., AAAI 2018] operating on 3D poses and shelf positions.
4. Camera view selection — the two cameras with the best visibility of the person and their hands are selected per interval.
5. Frame subsampling — at most 32 frames are retained per clip using a model that scores frames by hand velocity and acceleration, prioritizing frames with significant movement.
6. Anonymization — facial blurring, timestamp anonymization to epoch 1970-01-01T00:00:00, and removal/blurring of store name references.

Who are the source data producers?

Real customers visiting operational convenience stores. All shoppers were given notice of recordings and signed terms of service with Standard AI before data collection. No actors or scripted interactions were used.

Annotations

Annotation process

Annotation followed a two-step process:

1. Binary classification and quality labeling — annotators marked each segment as containing an interaction or not, and flagged data quality issues (bad camera view, low resolution, too few frames, pose errors). A model-in-the-loop strategy iterated three times, reviewing the 10% most disagreeing samples after each round.
2. Spatio-temporal fine-grained labeling — annotators precisely marked the temporal start and end of each individual interaction and placed a 2D spatial point on the exact location where the subject's hand contacts the product, for both camera views. In multi-person clips, a red dot on the subject's head in every frame guided annotators to label only the designated individual.

Action categories are mutually exclusive per interaction: take (picking up an item), put (placing an item back), and touch (hand contact without taking or placing).

Who are the annotators?

[More Information Needed]

Personal and Sensitive Information

All individuals in the dataset have been anonymized:

• Faces are blurred in every video frame
• All timestamps are reset to epoch 1970-01-01T00:00:00
• Store name references are removed or blurred
• Person identifiers across splits are withheld; shopper identity is not released

Citation

BibTeX:

@inproceedings{mazzini2025retailaction,
  title={RetailAction: Dataset for Multi-View Spatio-Temporal Localization of Human-Object Interactions in Retail},
  author={Mazzini, Davide and Raimondi, Alberto and Abbate, Bruno and Fischetti, Daniel and Woollard, David M.},
  booktitle={ICCV Retail Vision Workshop},
  year={2025}
}

APA:

Mazzini, D., Raimondi, A., Abbate, B., Fischetti, D., & Woollard, D. M. (2025). RetailAction: Dataset for multi-view spatio-temporal localization of human-object interactions in retail. ICCV Retail Vision Workshop. https://doi.org/10.1109/iccvw69036.2025.00250

More Information

• Source repository: https://huggingface.co/datasets/standard-cognition/RetailAction
• Competition: PRAW @ WACV 2027 — Track 3: TV-STAL
• Contact: {davide, bruno, david.woollard}@standard.ai

Dataset Card Authors

Harpreet Sahota

Dataset Card Contact

[More Information Needed]