AI-Powered
Plant Disease
Detection
Detect plant diseases in seconds. 23 pretrained models — from lightweight MobileNetV2 to research-grade Recursive Additive Attention — all trained on the PlantVillage 38-class dataset.
Pretrained Models
CNN, ViT, Swin, ConvNeXt, RegNet
Peak Accuracy
Recursive Additive Attention v1
Plant Disease Classes
PlantVillage dataset
Supported Crops
Apple, Tomato, Grape & more
How It Works
From leaf to diagnosis
Three steps, under three seconds. The same pipeline powers both the Android app and this website.
Upload a Leaf Image
Take a photo with the Android app or drag & drop an image on this website. Accepts JPG, PNG, or WebP.
AI Model Inference
Your image is sent to a FastAPI backend powered by plantdoc-predictor. Choose from 23 pretrained models.
Instant Diagnosis
Get the disease name, confidence score, crop type, and whether the plant is healthy — in under 3 seconds.
Features
Everything you need for plant AI
Whether you are a researcher, developer, or farmer — PlantDoc AI has a workflow for you.
23 AI Models
Model ZooFrom fast MobileNetV2 to research-grade Recursive Additive Attention. Pick your accuracy/speed tradeoff. Models auto-download on first use.
Android App
MobileNative React Native app with camera scan, offline model selection, scan history, Firebase sync, and disease info cards.
Smart Leaf Guard
SafetyCLIP-based guard rejects non-leaf images before inference. Prevents false positives on random photos.
Grad-CAM XAI
ExplainabilityExplainable predictions — a heatmap overlay shows exactly which leaf regions drove the model's decision.
Batch Mode
AutomationRun inference over entire folders or lists. Export results to CSV or JSON. Ideal for research pipelines.
Open REST API
APIThe same FastAPI backend powers the app and this website. Drop into any stack with a single POST request.
Model Zoo
23 pretrained architectures
CNN classics, Vision Transformers, Swin, ConvNeXt, and RegNet — all trained on the PlantVillage 38-class dataset.
Recursive Additive Attention v1
99.7%
KerasDenseNet169 v1
99.68%
KerasConvNeXt Small v1
99.5%
Keras| Rank | Model | Accuracy | Framework | Speed | Description |
|---|---|---|---|---|---|
| 🥇 | rec_add_attention_v1 | 99.7% | Keras | slow | Top accuracy — research model |
| 🥈 | densenet169_v1recommended | 99.68% | Keras | medium | Best classical CNN |
| 🥉 | convnext_small_v1 | 99.5% | Keras | medium | High accuracy + efficient |
| 4 | convnext_tiny_v1 | 99.23% | Keras | fast | Lightweight ConvNeXt |
| 5 | convnext_base_v1 | 99.1% | Keras | medium | ConvNeXt base |
| #6 | swin_tiny_patch4_window7 | 99.1% | PyTorch | medium | Lightweight Swin Transformer |
| 6 | swin_base_patch4_window7 | 99.1% | PyTorch | slow | Swin Transformer |
| #8 | vit_tiny_16_v1 | 99.1% | PyTorch | fast | Lightweight Vision Transformer |
| #9 | vit_small_16_v1 | 99.1% | PyTorch | medium | Small Vision Transformer |
| 7 | vit_base_16_v1 | 99.1% | PyTorch | slow | ViT base |
| #11 | vit_large_16_v1 | 99.1% | PyTorch | slow | ViT large — max capacity |
| 8 | vgg19_v1 | 98.98% | Keras | medium | Deep VGG |
| #13 | densenet121_v1 | 98.68% | Keras | fast | Efficient DenseNet |
| #14 | inceptionv3_v1 | 98.2% | Keras | medium | Multi-scale InceptionV3 |
| #15 | resnet50_v1 | 97.8% | Keras | fast | Residual learning CNN |
| #16 | efficientnetb50_v1 | 97.8% | Keras | fast | EfficientNet balanced |
| #17 | densenet210_v1 | 97% | Keras | medium | Very deep DenseNet |
| #18 | mobilenetv2_v1 | 96.8% | Keras | fast | Fastest — mobile optimized |
| #19 | vgg16_v1 | 96.8% | Keras | medium | Classic CNN baseline |
| #20 | alexnet_v1 | 92.8% | Keras | fast | Historical baseline |
Live Demo
Try it right now
Upload a leaf photo and get an AI diagnosis instantly. Uses the same backend as the Android app — powered by plantdoc-predictor.
Drag & drop a leaf image
or click to browse · JPG, PNG, WebP · max 10 MB
Results typically arrive in under 3 seconds
Your diagnosis will appear here
Upload a leaf image and click Analyze
Installation
One command to get started
pip install plantdoc-predictorfrom plantdoc_predictor import Predictor
predictor = Predictor(model_name="densenet169_v1")
result = predictor.predict("leaf.jpg")
print(result)
# {
# 'model': 'densenet169_v1',
# 'label': 'Apple___Apple_scab',
# 'confidence': 0.984,
# 'plant': 'Apple',
# 'disease': 'Apple scab',
# 'is_healthy': False
# }Research
Peer-reviewed science
Recursive Additive Attention for Plant Disease Classification
Subham Divakar, Rojalina Priyadarshini
A novel CNN architecture incorporating Recursive Additive Attention mechanisms for enhanced feature interaction. Achieves 99.70% accuracy on the PlantVillage 38-class plant disease dataset — the highest in the PlantDoc Model Zoo.
rec_add_attention_v1Upcoming research models
Multimodal models for plant disease detection have been presented at several conferences and are targeted for inclusion in a future PlantDoc Model Zoo release.
🌿 PlantDoc AI
AI-powered disease detection
Tap to scan leaf
4
Total
3
Healthy
1
Diseased
Active Model
densenet169_v1
99.68% accuracy
Android App
Scan leaves anywhere
The PlantDoc AI Android app puts 20 AI models in your pocket. Scan in the field, get an instant diagnosis with disease info, and sync your history to the cloud.
- Camera scan or gallery upload
- 20 AI models selectable on-device
- Full scan history with cloud sync
- Disease info: symptoms, causes, treatment
- Firebase authentication & cloud backup
- Offline-ready model selection