casiatao/InfMix

---
license: apache-2.0
tags:
- infrared
- image
- pretraining
size_categories:
- 1M<n<10M
---

<h1 align="center"> UNIP: Rethinking Pre-trained Attention Patterns for Infrared Semantic Segmentation </h1>

<p align="center">
<a href='https://arxiv.org/abs/2502.02257'>
<img src='https://img.shields.io/badge/Arxiv-2502.02257-A42C25?style=flat&logo=arXiv&logoColor=A42C25'></a>
<a href='https://huggingface.co/casiatao/UNIP'>
<img src='https://img.shields.io/badge/%F0%9F%A4%97%20Model-UNIP-yellow'></a>
</p>

Code: https://github.com/casiatao/UNIP

## 📖 Introduction
This repository releases the InfMix dataset, a pre-training dataset consisting of 541,088 infrared images collected from 23 sub-datasets.

Scripts for unzipping
```
#!/bin/bash

LOGFILE="unzip_errors.log"
: > "$LOGFILE" # Clear the log file

files=(*.zip *.z??)
declare -A processed

for file in "${files[@]}"; do
[[ -f "$file" ]] || continue

# Extract the base name without extension, e.g., a.zip -> a, a.z01 -> a
base="${file%%.*}"

# Skip if already processed
if [[ -n "${processed[$base]}" ]]; then
continue
fi
processed[$base]=1

# If the target directory already exists, skip extraction
if [[ -d "$base" ]]; then
echo "Directory $base exists, skipping unzip of $base.zip"
continue
fi

mkdir -p "$base"

# Extract using 7z (quiet mode), redirect errors to log, and check success
if ! 7za x -y "$base.zip" -o"$base" >>"$LOGFILE" 2>&1; then
echo "[$(date '+%F %T')] Error: Failed to 7z extract $base.zip" >> "$LOGFILE"
echo "Skipping $base.zip due to error."
rm -rf "$base"
continue
fi
done

echo "Unzip task completed. Please check $LOGFILE for error logs."
```

## Citation
If you find this repository helpful, please consider giving a like ❤️ and citing:
```bibtex
@inproceedings{
zhang2025unip,
title={{UNIP}: Rethinking Pre-trained Attention Patterns for Infrared Semantic Segmentation},
author={Tao Zhang and Jinyong Wen and Zhen Chen and Kun Ding and Shiming Xiang and Chunhong Pan},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=Xq7gwsnhPT}
}
```

license: apache-2.0
tags:
- 红外
- 图像
- 预训练
size_categories:
- 100万<样本数<1000万
---

<h1 align="center"> UNIP：重新思考红外语义分割的预训练注意力模式</h1>

<p align="center">
<a href='https://arxiv.org/abs/2502.02257'>
<img src='https://img.shields.io/badge/ArXiv-2502.02257-A42C25?style=flat&logo=arXiv&logoColor=A42C25'></a>
<a href='https://huggingface.co/casiatao/UNIP'>
<img src='https://img.shields.io/badge/%F0%9F%A4%97%20模型-UNIP-yellow'></a>
</p>

代码仓库：https://github.com/casiatao/UNIP

## 📖 数据集介绍
本仓库发布了InfMix数据集，这是一款预训练数据集，包含从23个子数据集中采集的541088张红外图像。

### 解压脚本
bash
#!/bin/bash

# 定义日志文件路径
LOGFILE="unzip_errors.log"
# 清空日志文件
: > "$LOGFILE"

files=(*.zip *.z??)
declare -A processed

for file in "${files[@]}"; do
[[ -f "$file" ]] || continue

# 提取不带扩展名的基础文件名，例如 a.zip → a，a.z01 → a
base="${file%%.*}"

# 若已处理过该文件则跳过
if [[ -n "${processed[$base]}" ]]; then
continue
fi
processed[$base]=1

# 若目标目录已存在，则跳过当前解压任务
if [[ -d "$base" ]]; then
echo "目录 $base 已存在，跳过解压 $base.zip"
continue
fi

mkdir -p "$base"

# 使用7za静默模式解压，将错误输出重定向至日志文件，并校验解压结果
if ! 7za x -y "$base.zip" -o"$base" >>"$LOGFILE" 2>&1; then
echo "[$(date '+%F %T')] 错误：无法解压 $base.zip" >> "$LOGFILE"
echo "因解压失败，跳过 $base.zip。"
rm -rf "$base"
continue
fi
done

echo "解压任务已完成，请查看 $LOGFILE 获取详细错误日志。"

## 📚 引用规范
若本仓库对你的研究工作有所助益，请点赞❤️并引用如下文献：
bibtex
@inproceedings{
zhang2025unip,
title={{UNIP}: Rethinking Pre-trained Attention Patterns for Infrared Semantic Segmentation},
author={Tao Zhang and Jinyong Wen and Zhen Chen and Kun Ding and Shiming Xiang and Chunhong Pan},
booktitle={The Thirteenth International Conference on Learning Representations},
year={2025},
url={https://openreview.net/forum?id=Xq7gwsnhPT}
}