MTIL - Mini tomato image library

1. What's MTIL? MTIL - Mini Tomato Image Library is a dataset with images of Mila c.v, taken in a greenhouse in Agricultural Engineering School at University of Campinas, Brazil. 2. What's it's purpose? It was create to be used in deep learning segmentation tasks. 3. Where's this images came from? The images was taken in a greenhouse located at 22°49'07.1"S 47°03'40.8"W at 635 meters from the sea level, between 07/september and 29/october of 2019. 4. How this images was taken? The images was taken by digital cameras mounted in tripods inside greenhouse, about 70 cm from the tomato clusters. The camera is a generic 5 MP full HD OV5647 sensor, connected to a raspberry pi zero wifi. 5. How the figures are organized? Inside project there are two main directories: clipped: Contains 385 images clipped to 1200x1200 pixels, trying to center the foreground cluster. Each image here has a correspondent mask in mask. mask: Contains 385 binary[ representations of images in clipped. Here the class "tomato" is represented as white and "non-tomato" as black. 6. How these figures are named: Figures was named as c_YYYY-MM-DD_hh_mm_bb.jpg Where c is the color of flash used. may be w for white, r for red, g for green or b for blue; YYYY-MM-DD is respectively the year, month and day when the picture was taken; hh:mm or hh_mm similarly, is the hour and minutes when picture was taken, in 24h format; bb is the bunch identification. Besides it was built for binary, image itself has three channels. The image editor was not perfect to make the labelled images, near silhouette border are some values slightly above 255.

1. 什么是MTIL？MTIL即迷你番茄图像库（Mini Tomato Image Library），是一组针对Mila品种番茄的图像数据集，拍摄于巴西坎皮纳斯大学农业工程学院的温室中。
2. 该数据集的用途是什么？本数据集的创建初衷为深度学习分割任务。
3. 这些图像源自何处？所有图像拍摄于坐标为南纬22°49'07.1"、西经47°03'40.8"、海拔635米的温室中，拍摄时段为2019年9月7日至2019年10月29日。
4. 这些图像是如何拍摄的？图像由安装在温室三脚架上的数码相机拍摄，相机与番茄果簇的距离约为70厘米。所用相机为通用型5 MP全高清OV5647传感器，连接至树莓派Zero WiFi。
5. 数据集的文件结构如何组织？本项目包含两个核心目录：
- `clipped`（裁剪图像目录）：内含385张裁剪为1200×1200像素的图像，旨在将前景番茄果簇置于画面中央。该目录下每张图像均对应`mask`目录中的一张标注掩码。
- `mask`（掩码目录）：内含385张与`clipped`目录中图像对应的二值掩码图像。其中"番茄"类别以白色表示，"非番茄"类别以黑色表示。
6. 图像的命名规则是什么？图像采用`c_YYYY-MM-DD_hh_mm_bb.jpg`格式命名，各字段含义如下：
- `c`：闪光灯颜色标识，取值可为w（白色）、r（红色）、g（绿色）或b（蓝色）；
- `YYYY-MM-DD`：依次代表图像拍摄的年、月、日；
- `hh:mm`或`hh_mm`：采用24小时制的拍摄时、分；
- `bb`：果簇编号。
尽管本数据集专为二值分割任务构建，但原始图像本身为三通道彩色图像。由于用于生成标注图像的图像编辑工具存在缺陷，在轮廓边界附近存在少量略高于255的像素值。

搜集汇总

背景与挑战

背景概述

MTIL是一个专门用于深度学习图像分割任务的番茄图像数据集，包含385张1200x1200像素的裁剪图像及对应的二进制掩码，图像于2019年在巴西坎皮纳斯大学的温室环境中标准化采集。该数据集适用于农业计算机视觉研究，特别是番茄目标分割，采用CC BY-NC 4.0许可证。