iLearn-Lab/MM26-ADGNet-AITIR-Text

1---2license: apache-2.03language:4  - en5pretty_name: AITIR Asymmetric Text Annotations6tags:7  - infrared-small-target-detection8  - multimodal9  - vision-language10  - text-annotations11 12---13 14 15<h1>📝 AITIR: Asymmetric Image-Text Infrared Text Annotations</h1>16 17<p>18  Official asymmetric text annotations for infrared small target detection.19</p>20 21 22<p>23  <a href="https://github.com/iLearn-Lab/MM26-ADGNet">24    <img src="https://img.shields.io/badge/GitHub-MM26--ADGNet-black?logo=github" alt="GitHub">25  </a>26  <a href="<paper-link>">27    <img src="https://img.shields.io/badge/ACM%20MM-2026-blue" alt="ACM MM 2026">28  </a>29</p>30 31---32 33## 📖 Description34 35This repository provides the official asymmetric text annotations used to construct the **Asymmetric Image-Text Infrared (AITIR)** dataset introduced in:36 37**ADGNet: Asymmetric Dual-text Guided Network for Infrared Small Target Detection**, accepted by **ACM Multimedia 2026**.38 39The released annotations cover three infrared small target detection datasets:40 41- **IRSTD-1K**42- **NUDT-SIRST**43- **SIRST**44 45This repository contains text annotations only. The original infrared images and ground-truth masks are not included.46 47---48 49## 🏷️ Annotation Design50 51Each infrared image is associated with two asymmetric text prompts.52 53### Fixed Target Prompt54 55A concise and image-independent target prompt is shared across all images:56 57```text58an infrared image featuring one or multiple target59```60 61### Detailed Background Prompt62 63Each image is assigned an image-dependent background prompt following the template:64 65```text66an infrared [S] image with [C]67```68 69where:70 71- `[S]` describes the global infrared scene.72- `[C]` describes local structures, thermal clutter, and potential distractors.73 74Example:75 76```text77an infrared cloudy sky image with faint dark cloud structures and bright illuminated building tops78```79 80---81 82## 📊 Annotation Coverage83 84|  Dataset   | Fixed Target Prompt | Detailed Background Prompt |85| :--------: | :-----------------: | :------------------------: |86|  IRSTD-1K  |          ✓          |             ✓              |87| NUDT-SIRST |          ✓          |             ✓              |88|   SIRST    |          ✓          |             ✓              |89 90---91 92## 📁 Repository Structure93 94The asymmetric text annotations are organized by dataset and data split:95 96```text97MM26-ADGNet-AITIR-Text/98├── IRSTD-1K/99│   └── text/100│       ├── train_fg.json101│       ├── train_bg.json102│       ├── test_fg.json103│       └── test_bg.json104├── NUDT-SIRST/105│   └── text/106│       ├── train_fg.json107│       ├── train_bg.json108│       ├── test_fg.json109│       └── test_bg.json110└── SIRST/111    └── text/112        ├── train_fg.json113        ├── train_bg.json114        ├── test_fg.json115        └── test_bg.json116```117 118---119 120## 🧾 Annotation Files121 122Each dataset contains four JSON annotation files:123 124| File            | Description                                        |125| :-------------- | :------------------------------------------------- |126| `train_fg.json` | Target prompts for the training split              |127| `train_bg.json` | Detailed background prompts for the training split |128| `test_fg.json`  | Target prompts for the testing split               |129| `test_bg.json`  | Detailed background prompts for the testing split  |130 131Here, `fg` denotes the foreground or target prompt, while `bg` denotes the detailed background prompt.132 133---134 135## 🚀 Usage136 137Download the annotations and place the corresponding JSON files under the `text/` directory of each original dataset:138 139```text140datasets/141├── IRSTD-1K/142│   ├── images/143│   ├── masks/144│   ├── img_idx/145│   └── text/146│       ├── train_fg.json147│       ├── train_bg.json148│       ├── test_fg.json149│       └── test_bg.json150├── NUDT-SIRST/151│   ├── images/152│   ├── masks/153│   ├── img_idx/154│   └── text/155│       ├── train_fg.json156│       ├── train_bg.json157│       ├── test_fg.json158│       └── test_bg.json159└── SIRST/160    ├── images/161    ├── masks/162    ├── img_idx/163    └── text/164        ├── train_fg.json165        ├── train_bg.json166        ├── test_fg.json167        └── test_bg.json168```169 170> [!NOTE]171> The JSON entries correspond to the samples in the respective training or testing split. Please keep the provided filenames and directory structure unchanged when using them with the official ADGNet implementation.172 173## ⚠️ Notes174 175- This repository provides text annotations only.176- The original infrared images and ground-truth masks must be obtained separately.177- The annotations are written in English.178- Users must comply with the licenses and terms of use of the original datasets.179- Annotation filenames must match the corresponding infrared image filenames.180 181---182 183## 📚 Citation184 185If you find this project useful in your research, please consider citing our paper:186 187```bibtex188 189```190 191Please also consider checking out and citing our other related work:192 193```bibtex194 195```