CCMT Crop Pest and Disease Dataset (Tomato)

The tomato portion of the CCMT crop pest and disease dataset, mirrored from Mendeley Data (bwh3zbpkpv, v1) and described in Mensah et al., Data in Brief 49 (2023) 109306. The images are field photographs of tomato leaves taken on farms in Ghana and validated by plant virologists, each filed under one of five condition classes. The source covers four crops (cashew, cassava, maize, tomato) in a raw set and a much larger augmented set; this project holds the raw tomato images, one copy per distinct image, with the class recoverable from the file path. Cashew, cassava and maize can follow in the same project.

Contents

Images	5,325 raw JPEGs (of 5,805 in the source's Raw Data/CCMT Dataset/Tomato/ folders; see Conversion notes)
Labels	5,325 image-level classifications: one annotation per file carrying its class label (no geometry), plus the class in the path as images/Tomato/<class>/<file>; there are no region annotations
Classes	5: healthy (463), leaf blight (1,285), leaf curl (511), septoria leaf spot (2,294), verticulium wilt (772; the source's spelling of verticillium wilt)
Dataset	ccmt-crop-pest-v1
Split	None. The source's train/test split exists only in the augmented set, which is not included
Imaging	Colour photographs of tomato leaves, mostly single detached leaves laid on a plain light background and some still on the plant, 400x400 pixels, 8-bit RGB JPEG

The labels are image-level: a file's class is its folder, exactly as in the source, and each file carries exactly one annotation with that class label and no shape. Nothing was drawn or inferred, so there are no bounding boxes or polygons to export.

Using it

Browse the files by folder to see one class at a time. To download everything (free account required):

• ZIP: use the Download button on this project.
• Python SDK: pip install datatorch, then:

datatorch login
datatorch pull datatorch/ccmt-crop-pest

After pulling, the class of every image is the second-to-last path component (images/Tomato/<class>/<file>).

Conversion notes

• Download method: the source's whole-dataset zip (8.4 GB, 130,472 entries) was read with HTTP range requests: first the ZIP64 central directory, then the contiguous byte span holding the entries under Dataset for Crop Pest and Disease Detection/Raw Data/CCMT Dataset/Tomato/. Every entry was inflated and its CRC-32 verified. Image bytes are unchanged from the source; nothing was re-encoded.
• Label import: the five labels are the source's class folder names, spelled as in the source. Each of the 5,325 files was given one annotation with the label of its images/Tomato/<class>/ folder, through the GraphQL API (createAnnotations with an empty sources list), on 2026-10-27, and every file was then re-read to confirm it has exactly one annotation whose label matches its folder.
• Subset rule: exactly the 5,805 entries under that Tomato/ prefix. The augmented set and the other three crops were not downloaded.
• Excluded, 25 files that do not decode (Pillow 12: broken data stream, truncated file, or no JPEG header): healthy442_, healthy443_, healthy76_, healthy77_, leaf blight1232_, leaf blight386_, leaf blight470_, leaf blight471_, leaf blight558_, leaf blight644_, leaf blight645_, leaf blight72_, leaf blight813_, leaf blight814_, leaf blight8_, leaf blight997_, leaf blight998_, leaf curl186_, leaf curl270_, leaf curl271_, leaf curl353_, leaf curl354_, leaf curl439_, leaf curl43_, verticulium wilt403_ (all .jpg).
• Excluded, 454 duplicate copies: 454 pairs of files are byte-identical (SHA-256). One copy per pair was kept, the lexicographically first path. 453 of these pairs sit inside one class; the remaining pair, leaf blight/leaf blight1315_.jpg and septoria leaf spot/septoria leaf spot47_.jpg, carries two different labels, so both copies were left out as ambiguous. Net: 5,805 - 25 - 453 - 2 = 5,325 images.
• Discrepancies in the source: the Mendeley record describes 5,435 raw tomato images, but the zip contains 5,805. The class folder is spelled verticulium wilt (verticillium); the spelling is kept so paths match the source. Filenames end in an underscore (for example healthy100_.jpg), kept as is. 447 of the septoria leaf spot files carry a random six-character suffix in the source (for example septoria leaf spot2190_-FjP09t.jpg); those names come from the source and were not changed here.

License and citation

CC BY 4.0. Verified on the Mendeley Data record (licence badge and JSON-LD metadata both point to creativecommons.org/licenses/by/4.0). Attribution is required, and changes must be indicated: this copy is a subset with the exclusions listed above. If you use this data, please cite:

Mensah, P.K. et al. CCMT: Dataset for crop pest and disease detection. Data in Brief 49 (2023) 109306. https://doi.org/10.1016/j.dib.2023.109306

Source: Mendeley Data, bwh3zbpkpv v1, DOI 10.17632/bwh3zbpkpv.1.

Related

• More agriculture datasets on DataTorch
• All public datasets