README.md

metadata

license: cc-by-4.0
task_categories:
  - video-classification
tags:
  - thermal
  - action-recognition
  - human-activity-recognition
size_categories:
  - n<1K

OpenThermalAction - Sample Data (Balcony Environment)

This repository provides example thermal image sequences for action recognition captured in a balcony environment:

• Single-person sports activities (Session 1): One subject performing various sports actions
• Multi-person daily interactions (Session 2): One pair of subjects performing interactive activities

Structure

ota-balcony/
├── train/
│   ├── session_1/        # Session 1: Single-person sports
│   │   └── sub_1/
│   │       └── thermal/
│   │           ├── 1_1_0/    # Action sequences as image folders
│   │           ├── 1_2_0/
│   │           └── ...
│   └── session_2/        # Session 2: Multi-person interactions
│       └── exp_1/
│           └── thermal/
│               ├── 1/        # Action sequences as image folders
│               ├── 2/
│               └── ...
├── annotations_train.txt
└── annotations_val.txt

Each folder contains sequential thermal images representing one action clip.

Note: Action clips with suffixes like 1_1 and 1_2 represent different camera views of the same video.

Annotation Format

Each line in annotations_train.txt or annotations_val.txt contains:

path_to_sequence label

Example:

session_1/sub_1/thermal/1_1 0
session_1/sub_1/thermal/2_1 1
session_2/exp_1/thermal/1 15

Label Mapping

• Session 1: Label = action_number - 1 (classes 0-14)
• Session 2: Label = action_number + 14 (classes 15-27)