Datasets
On this webpage, you will find a selection of datasets that we consider suitable for applications related to the GoodBrother COST Action. If you spot any errors or would like to recommend an additional dataset, please get in touch.
Video datasets
- Action Recognition in the Dark (ARID)
- ActivityNet
- AVA (Atomic Visual Actions)
- CAD-120
- Charades
- Charades-Ego
- Composable activities dataset
- EgoGesture
- EgoHands
- EgoProceL
- EPIC-KITCHENS-100
- EPIC-KITCHENS-55
- First-Person Hand Action Benchmark
- G3D (Gaming 3D Dataset)
- HAA500 (Human-Centric Atomic Action Dataset)
- HACS (Human Action Clips and Segments)
- HDM05
- HMDB51
- HOMAGE – Home Action Genome
- IfAct (Identifying Human Actions Visible in Online Vlogs)
- InteractADL
- Jester (Gesture Recognition)
- JHMDB (Joint-annotated Human Motion Data Base)
- Kinetics-600
- Kinetics-700
- KTH Action dataset
- LIRIS human activities dataset
- MECCANO
- Metaphorics
- MMAct
- MPII Cooking 2 Dataset
- MultiTHUMOS
- NTU RGB+D
- NTU RGB+D 120
- ODIN – OmniDirectional Indoor Dataset
- OREBA (Objectively Recognizing Eating Behaviour and Associated Intake)
- PA-HMDB51 (Privacy Annotated HMDB51)
- Penn Action
- PKU-MMD
- Skeletics 152
- Skeleton-Mimetics
- Something-Something V1
- Something-Something V2
- THUMOS14
- Toyota Smarthome Dataset
- TSU (Toyota Smarthome Untrimmed)
- TUM Kitchen
- TVSeries Dataset
- UCF101 Human Actions dataset
- UTD-MHAD
Audio datasets
- CMU Multimodal Opinion Sentiment and Emotion Intensity (CMU-MOSEI)
- CREMA-D
- DCASE 2016
- DESED (Domestic environment sound event detection)
- EPIC-SOUNDS
- IEMOCAP (The Interactive Emotional Dyadic Motion Capture)
- Kinetics-Sound
- LSSED
- RAVDESS (Ryerson Audio-Visual Database of Emotional Speech and Song)
- VGG-Sound
- VocalSound
- WildDESED (Wild Domestic Environment Sound Event Detection)
Featured dataset
Members of the GoodBrother COST Action have acquired the OmniDirectional INdoor Dataset – ODIN to capture and analyse Activities of Daily Living (ADLs) in indoor environments. It utilises overhead omnidirectional cameras alongside multiple synchronised data modalities to advance research in applications related to AAL environments. The included modalities feature omnidirectional RGB images from overhead cameras, lateral camera data in RGB and infrared (IR) formats, and egocentric videos captured with chest-mounted cameras. Additionally, physiological signals such as heart rate and physical activity levels are collected, along with accelerometer data from wearable devices. The dataset also incorporates three-dimensional models of the environments obtained through depth scans.
The recordings were conducted in real indoor environments, including kitchens, bathrooms, bedrooms, and living rooms, aiming to reflect typical daily living conditions. This approach captures occlusions caused by furniture and objects present in the spaces, adding realism and complexity to the data. ODIN is the first resource of its kind to provide data from omnidirectional overhead views, making it essential to address the scarcity of similar resources.
