Bacterial Colony Images on Multiple Media (10-class dataset)
收藏资源简介:
# Bacterial Colony Images on Multiple Media (10-class dataset) This repository provides **documentation, metadata, scripts, and a lightweight GitHub subset** of a bacterial colony image dataset organized by **species × culture medium** (10 classes). ✅ **Full dataset (recommended for research use):** hosted on **Zenodo** (~9.2 GB) ✅ **GitHub subset:** a reduced version (~1.9 GB) containing images from **all 10 classes**, intended for quick access, preview, and reproducible folder structure (not the complete set). --- ### Zenodo (full archival dataset)- Full-resolution dataset (**~9.2 GB**) intended for experiments and publications.- DOI-based, citable, versioned archive. --- ## Quick facts (Full Zenodo release) - **Total images (full):** 2317 - **File formats:** JPG, PNG - **Classes:** 10 (4 species across multiple media) - **Full dataset size:** ~9.2 GB (Zenodo) - **GitHub subset size:** ~1.9 GB (this repo; not complete) - **Imaging protocol (paper):** fixed ≥16 MP camera, controlled lighting, ~30 cm standoff; consistent framing and preprocessing. --- ## Class taxonomy (species × medium) — Full dataset counts (Zenodo) | Species | Medium (folder name) | Images (full) ||---|---|---:|| E_Coli | E_Coli on EMB agar medium | 203 || E_Coli | E_Coli on MacConkey_Agar medium | 203 || E_Coli | E_Coli on Nutrients agar medium | 216 || Salmonella | Salmonella on XLD agar medium | 211 || Salmonella | Salmonella on MacConkey agar medium | 215 || Salmonella | Salmonella on Nutrients agar medium | 294 || Enterococcus | Enterococcus on Slantz_and_Bartley agar medium | 208 || Enterococcus | Enterococcus on Nutrients agar medium | 285 || Staphylococcus | Staphylococcus on MSA agar medium | 175 || Staphylococcus | Staphylococcus on Nutrient Agar | 307 || **Total** | | **2317** | > Note: The **GitHub subset contains fewer images than the table above** (subset is ~1.9 GB). > The table reflects the **full Zenodo release**. --- Description_file/conversations.jsonl is a JSON Lines annotation file (one JSON object per line) that links each colony image to multiple natural-language question–answer pairs in an instruction/VQA format—each record contains an id, an image filename (matching the corresponding file under Images/...), and a conversations array with a human question and an assistant answer (keys like {"from":"human","value":...} and {"from":"gpt","value":...}); in the current subset it includes 16,501 Q/A items covering 592 images, with questions designed to capture colony morphology and diagnostic cues (e.g., colour, size range, margin, elevation, surface texture, opacity), growth/distribution patterns (crowding, isolation suitability, approximate count, purity/contamination hints), and medium-specific indicators (e.g., lactose fermentation cues, EMB sheen, swarming/motility, hemolysis), so it can be used to (i) fine-tune or benchmark vision-language models for colony interpretation, (ii) build an interactive teaching/QA assistant for microbiology lab training, (iii) generate standardized captions/attribute labels for weak supervision, retrieval, or dataset search, and (iv) reproduce the exact prompts/answers used in downstream experiments by grouping records by image and pairing them with the corresponding plate image files in the GitHub subset or Zenodo archive.
# 多培养基细菌菌落图像数据集(10分类数据集) 本仓库提供了一套按「物种×培养基」分类的细菌菌落图像数据集的文档、元数据、脚本文件及轻量化GitHub子集,该数据集共包含10个类别。 ✅ **完整数据集(推荐用于科研)**:托管于**Zenodo**,大小约9.2 GB;✅ **GitHub子集**:精简版本,大小约1.9 GB,包含全部10个类别的图像,用于快速预览、获取及复现文件夹结构(非完整数据集)。 --- ### Zenodo(完整存档数据集) - 全分辨率数据集,大小约9.2 GB,适用于实验研究与论文发表。 - 基于DOI(数字对象唯一标识符)的可引用、带版本控制的存档资源。 --- ## 核心参数(Zenodo完整数据集版本) - **总图像数(完整数据集)**:2317张 - **文件格式**:JPG、PNG - **类别数**:10类(涵盖4个物种的多培养基样本) - **完整数据集大小**:约9.2 GB(Zenodo托管) - **GitHub子集大小**:约1.9 GB(本仓库提供,非完整数据集) - **成像规范(来自论文)**:使用≥1600万像素的固定相机、受控光照环境,拍摄距离约30 cm,采用统一的取景与预处理流程。 --- ## 分类体系(物种×培养基)——Zenodo完整数据集样本量统计 | 物种 | 培养基(文件夹名) | 完整数据集样本量 | |---|---|---:| | 大肠杆菌(E_Coli) | EMB琼脂培养基上的大肠杆菌 | 203 | | 大肠杆菌(E_Coli) | 麦康凯琼脂培养基上的大肠杆菌 | 203 | | 大肠杆菌(E_Coli) | 营养琼脂培养基上的大肠杆菌 | 216 | | 沙门氏菌(Salmonella) | XLD琼脂培养基上的沙门氏菌 | 211 | | 沙门氏菌(Salmonella) | 麦康凯琼脂培养基上的沙门氏菌 | 215 | | 沙门氏菌(Salmonella) | 营养琼脂培养基上的沙门氏菌 | 294 | | 肠球菌(Enterococcus) | Slantz-Bartley琼脂培养基上的肠球菌 | 208 | | 肠球菌(Enterococcus) | 营养琼脂培养基上的肠球菌 | 285 | | 葡萄球菌(Staphylococcus) | 甘露醇盐琼脂(MSA)培养基上的葡萄球菌 | 175 | | 葡萄球菌(Staphylococcus) | 营养琼脂培养基上的葡萄球菌 | 307 | | **总计** | | **2317** | > 注意:**GitHub子集的样本量少于上表统计值**(子集大小约1.9 GB),上表数据对应**Zenodo完整发布版本**。 --- `Description_file/conversations.jsonl` 是一份JSON Lines格式的标注文件(每行对应一个JSON对象),将每张菌落图像与多条自然语言问答对以指令/视觉问答(VQA)格式关联。每条记录包含ID、图像文件名(与`Images/`目录下的对应文件匹配),以及包含人类提问与助手回答的`conversations`数组(键值格式如`{"from":"human","value":...}`与`{"from":"gpt","value":...}`)。在当前子集中,该文件包含16501条问答对,覆盖592张图像,提问内容涵盖菌落形态与诊断特征(如颜色、尺寸范围、边缘形态、隆起程度、表面纹理、透明度)、生长/分布模式(菌落密集度、分离适用性、近似计数、纯度/污染提示)以及培养基特异性指标(如乳糖发酵特征、EMB琼脂光泽、游动/运动性、溶血特性)。该标注集可用于以下场景:(1) 微调或基准测试用于菌落解读的视觉语言模型;(2) 构建用于微生物学实验室培训的交互式教学问答助手;(3) 为弱监督学习、信息检索或数据集检索生成标准化图像描述与属性标签;(4) 通过按图像分组记录,并与GitHub子集或Zenodo存档中的对应平板图像文件配对,复现下游实验中使用的精确提示词与回答。



