ixrbhii/multicrop-disease-models

Multi-crop disease models (42 crops)

Leaf-photo models of the Multi-Crop Disease Decision Support System: two crop detectors that recognise which crop a leaf belongs to, and one disease model per crop for 42 crops. All are ConvNeXt-Small (timm, convnext_small.fb_in22k_ft_in1k fine-tuned), 224 × 224 RGB input.

They power Agri ▸ Diagnose crop disease in LULC Fetch, which downloads each model here the first time it's needed. The matching Q&A knowledge base is ixrbhii/crop-disease-qa.

Just one crop?

Every model also has its own repository, ready for timm (listed in the table below, and in the collection Multi-crop disease models (42 crops)):

import timm
model = timm.create_model("hf-hub:ixrbhii/tomato-disease-convnext", pretrained=True)   # labels: model.pretrained_cfg["label_names"]

Crop detectors: leaf-crop-detector-16-convnext and leaf-crop-detector-36-convnext. Or download one folder of this repository: hf download ixrbhii/multicrop-disease-models --include "Tomato/*".

Files

<Crop>/model.safetensors        disease model (half precision, ~100 MB)
<Crop>/config.json              arch, classes (output order), friendly labels, input size, normalisation, test scores
detectors/original/…            crop detector, 16 crops (99.8 % on 2,621 test photos)
detectors/new/…                 added-crops detector, 36 crops incl. Pepper, Raspberry, Sorghum, Squash (98.3 % on 6,167 test photos)
index.json                      every file with its size and SHA-256

Weights are stored in half precision (converted from the original float32 checkpoints; checked to give the same top-1 answer on sample leaf photos). Load them in float32 for CPU use.

Use

import json, timm, torch
from huggingface_hub import hf_hub_download
from safetensors.torch import load_file
from PIL import Image
import numpy as np

repo, crop = "ixrbhii/multicrop-disease-models", "Tomato"
cfg = json.load(open(hf_hub_download(repo, f"{crop}/config.json")))
net = timm.create_model(cfg["arch"], pretrained=False, num_classes=len(cfg["classes"]))
net.load_state_dict({k: v.float() for k, v in load_file(hf_hub_download(repo, f"{crop}/model.safetensors")).items()})
net.eval()

img = Image.open("leaf.jpg").convert("RGB").resize((224, 224), Image.BILINEAR)
x = (np.asarray(img, np.float32) / 255 - cfg["mean"]) / cfg["std"]
p = torch.softmax(net(torch.from_numpy(x.transpose(2, 0, 1)).float()[None]), 1)[0]
for i in p.argsort(descending=True)[:3]:
    print(cfg["labels"][i], f"{100 * p[i]:.1f} %")

Photo-first flow (as in the app): run detectors/original; if detectors/new names a crop outside Apple, Mango, Cashew, Cherry, Strawberry, Rose with at least 80 % confidence, use that instead. Refuse the photo when the crop detector is below 60 % or the disease model below 45 % sure, then run the crop's disease model.

Models and test results

Test accuracy is on held-out photos from the same datasets the models were trained on: real field photos score lower. Mulberry's model tells varieties, not diseases.

Crop	Own repository	Folder here	Classes	Test accuracy	Macro-F1	Test photos
Apple	apple-disease-convnext	Apple/	Apple Scab & Rust, Gray Spot, Apple Scab, Brown Spot, Alternaria Leaf Spot, Healthy, Marssonina Blotch, Root Rot / Collar Rot	99.2%	0.989	2,308
Apricot	apricot-disease-convnext	Apricot/	Blight, Healthy, Shot Hole	100.0%	1.000	277
Banana	banana-disease-convnext	Banana/	Bract Mosaic Virus Disease, Healthy, Insect Pest Damage, Moko Disease, Yellow Sigatoka Disease, Black Streak, Bunchy Top, Cordana Spot, Fruit Anthracnose, Panama Disease (Fusarium Wilt)	98.8%	0.985	748
Betel	betel-disease-convnext	Betel/	Bacterial Leaf Disease, Dried Leaf, Fungal Brown Spot, Healthy	100.0%	1.000	376
Bitter Gourd	bitter-gourd-disease-convnext	Bitter_gourd/	Downy Mildew, Fusarium Wilt, Healthy, Mosaic Virus, Nutrient Deficiency	94.9%	0.953	413
Black Pepper	black-pepper-disease-convnext	Black_Pepper/	Healthy, Leaf Blight, Yellow Mottle Virus	100.0%	1.000	84
Bottle Gourd	bottle-gourd-disease-convnext	Bottle_gourd/	Alternaria Leaf Blight, Anthracnose, Downy Mildew, Healthy, Mosaic Virus, Nutrient Deficiency	92.8%	0.901	428
Brinjal	brinjal-disease-convnext	Brinjal/	Begomovirus, Verticillium Wilt, Nutrient Deficiency, Shoot and Fruit Borer, MIT/EB Pest Damage, Healthy, Cercospora Spot, Phomopsis Fruit Rot	98.0%	0.976	587
Cashew	cashew-disease-convnext	Cashew/	Anthracnose, Gummosis, Healthy, Leaf Miner, Red Rust	97.4%	0.978	820
Cassava	cassava-disease-convnext	Cassava/	Healthy, Bacterial Blight, Brown Spot, Green Mite, Mosaic Virus	95.3%	0.955	721
Cherry	cherry-disease-convnext	Cherry/	Leaf Scorch, Healthy, Powdery Mildew, Brown Spot, Purple Spot, Shot Hole	99.4%	0.992	1,126
Chrysanthemum	chrysanthemum-disease-convnext	Chrysanthemum/	Bacterial Leaf Spot, Healthy, Septoria Leaf Spot	100.0%	1.000	372
Citrus (Orange / Lemon)	citrus-disease-convnext	Citrus/	Bacterial Blight, Black Spot, Canker, Dry Leaf, Healthy, Leaf Curl Virus, Sooty Mould, Spider Mites	99.5%	0.995	425
Coconut	coconut-disease-convnext	Coconut/	Bud Root Dropping, Bud Rot, Coconut Caterpillar Infestation (caterpillars), Coconut Caterpillar Infestation (leaflet damage), Gray Leaf Spot, Healthy, Leaf Rot, Stem Bleeding, Weligama Coconut Leaf Wilt — drying of leaflets, Weligama Coconut Leaf Wilt — flaccidity, Weligama Coconut Leaf Wilt — yellowing	99.9%	0.998	1,402
Coffee	coffee-disease-convnext	Coffee/	Berry Blotch, Cercospora Brown Eye Spot, Healthy, Leaf Miner, Phoma, Red Spider Mite, Rust	94.7%	0.809	506
Cucumber	cucumber-disease-convnext	Cucumber/	Angular Leaf Spot, Anthracnose, Bacterial Wilt, Downy Mildew, Gummy Stem Blight, Healthy, Nutrient Deficiency, Powdery Mildew	97.9%	0.971	819
Cucurbit (Pumpkin / Melon)	cucurbit-disease-convnext	Cucurbit/	Downy Mildew, Healthy, Leaf Curl, Mosaic Virus	99.3%	0.992	556
Custard Apple	custard-apple-disease-convnext	Custard_apple/	Anthracnose, Black Canker, Healthy, Stressed Plant, Diplodia Rot, Leaf Spot (leaves), Leaf Spot (fruit), Mealy Bug	99.4%	0.995	1,129
Fig	fig-disease-convnext	Fig/	Blight, Brown Spot, Healthy, Rust	99.5%	0.993	372
Grape	grape-disease-convnext	Grape/	Anthracnose, Bacterial Rot, Black Rot, Brown Spot, Downy Mildew, Esca Black Measles, Healthy, Leaf Blight, Mites, Powdery Mildew, Shot Hole	94.0%	0.942	1,461
Guava	guava-disease-convnext	Guava/	Healthy, Rust / Leaf Spot, Tea Mosquito Bug / Thrips, Anthracnose, Multiple Diseases, YLD	100.0%	1.000	1,345
Loquat	loquat-disease-convnext	Loquat/	Healthy, Leaf Spot	96.2%	0.961	212
Maize	maize-disease-convnext	Maize/	Common Rust, Fall Armyworm, Grasshopper Damage, Gray Leaf Spot, Healthy, Holcus Leaf Spot, Leaf Beetle, Lethal Necrosis, Streak Virus, Northern Leaf Blight, Smut	92.9%	0.936	1,281
Mango	mango-disease-convnext	Mango/	Anthracnose, Bacterial Canker, Cutting Weevil, Die Back, Gall Midge, Healthy, Powdery Mildew, Sooty Mould	99.8%	0.998	527
Mulberry	mulberry-variety-convnext	Mulberry/	ChiangMai60 (variety), RedKing (variety), WhiteKing (variety), BlackOodTurkey (variety), Taiwan Strawberry (variety), BlackAustralia (variety), Buriram60 (variety), Kamphaengsaeng42 (variety), TaiwanMeacho (variety), ChiangMaiBuriram60 (variety)	97.8%	0.977	671
Okra	okra-disease-convnext	okra/	Alternaria Leaf Spot, Cercospora Leaf Spot, Downy Mildew, Healthy, Leaf Curl Virus, Phyllosticta Leaf Spot	97.5%	0.977	399
Papaya	papaya-disease-convnext	Papaya/	Anthracnose, Bacterial Spot, Leaf Curl, Healthy, Mealybug, Mite Disease, Mosaic, Ringspot	97.1%	0.973	2,023
Peach	peach-disease-convnext	Peach/	Bacterial Spot, Brown Rot, Healthy, Leaf Curl	98.2%	0.974	284
Pear	pear-disease-convnext	Pear/	Black Spot, Fire Blight, Healthy, Leaf Spot, Slug	96.8%	0.972	631
Pomegranate	pomegranate-disease-convnext	pomrgranet/	Alternaria, Anthracnose, Bacterial Blight, Cercospora, Healthy	99.2%	0.991	638
Potato	potato-disease-convnext	Potato/	Early Blight, Healthy, Late Blight	96.0%	0.971	247
Rice	rice-disease-convnext	Rice/	Bacterial Blight, Bacterial Leaf Streak, Bacterial Panicle Blight, Blast, Brown Spot, Dead Heart, Downy Mildew, Healthy, Hispa, Leaf Scald, Narrow Brown Leaf Spot, Neck Blast, Sheath Blight, Tungro	93.5%	0.940	1,579
Ridge Gourd	ridge-gourd-disease-convnext	Ridge_gourd/	Healthy, Leaf Miner, Mosaic Virus, Nutrient Deficiency	91.0%	0.924	477
Rose	rose-disease-convnext	Rose/	Black Spot, Downy Mildew, Insect Hole, Healthy, Rust, Rose Slug (Sawfly), Yellow Mosaic Virus, Blight	99.3%	0.992	1,097
Snake Gourd	snake-gourd-disease-convnext	Snake_gourd/	Anthracnose, Healthy, Nutrient Deficiency, Yellowing	93.4%	0.940	152
Soybean	soybean-disease-convnext	Soybean/	Downy Mildew, Frogeye Leaf Spot, Healthy, Mosaic Virus, Rust, Septoria Brown Spot	91.7%	0.881	336
Strawberry	strawberry-disease-convnext	Strawberry/	Healthy, Leaf Scorch, Anthracnose	99.8%	0.995	459
Sugarcane	sugarcane-disease-convnext	Sugarcane/	Banded Chlorosis, Brown Rust, Brown Spot, Dried Leaves, Grassy Shoot, Healthy, Pokkah Boeng, Sett Rot, Smut, Yellow Leaf, Mosaic Virus	92.6%	0.899	689
Tea	tea-disease-convnext	Tea/	Healthy, Red Leaf Spot, Red Scab, Leaf Blight	100.0%	1.000	160
Tomato	tomato-disease-convnext	Tomato/	Bacterial Spot, Early Blight, Fusarium Wilt, Healthy, Late Blight, Leaf Miner, Leaf Mold, Mosaic Virus, Septoria Leaf Spot, Spider Mites, Target Spot, Verticillium Wilt, Yellow Leaf Curl Virus	93.4%	0.932	1,796
Walnut	walnut-disease-convnext	Walnut/	Anthracnose, Blotch, Gall Mite, Healthy, Shot Hole	99.7%	0.996	582
Watermelon	watermelon-disease-convnext	Watermelon/	Anthracnose, Downy Mildew, Healthy, Mosaic Virus	100.0%	1.000	300

Training data

• 16 crops (Apple, Banana, Brinjal, Cashew, Cherry, Coconut, Custard Apple, Guava, Mango, Mulberry, Okra, Papaya, Pomegranate, Rose, Strawberry, Watermelon) and the original crop detector: public leaf-disease photo datasets, including PlantVillage (CC0) and MangoLeafBD (CC-BY-4.0).
• 26 crops and the added-crops detector: LeafNet (CC-BY-4.0) and the authors' own field photos.

Limitations

• Leaves of crops the detectors don't know can be taken for a known crop; the confidence limits catch only part of them.
• Coffee's Cercospora brown eye spot is usually missed (few training photos).
• One leaf per photo, filling most of the picture, in daylight and in focus, works best.
• A diagnosis supports, but doesn't replace, a local agriculture expert.

Licence

CC-BY-4.0. Please credit "Multi-crop disease models, IXR (agritechixr)", and the training datasets above.