⌖
Every region holds information.
Segmentation helps reveal regions with possible signs of disease, directly on the leaf photo.
Highlighted regions · Example result
VISUAL INTELLIGENCE FOR RICE FARMING
Understand the leaf.
Find a clearer next step.
One photo to explore signs of rice disease. RiceLeaf helps you recognize possible symptoms and understand the regions that need attention.
No account neededPhotos stay in your browser4 supported diseases
Private by designAnalysis on your device
Beyond a prediction.See the regions highlighted by the model.
BUILT TO RECOGNIZE
Bacterial Blight✳Blast✳Brown Spot✳Tungro
SOLUTIONSSee more clearly.
See more clearly.
Understand more.
From visible symptoms to initial insights you can review with an agricultural adviser or expert.
Four possibilities.
One clear view.
Compare model scores for the four supported diseases.
Blast82,62%
Brown Spot16,37%
Example product result. Scores do not guarantee accuracy.Your photo.
Stays with you.
The model runs in your browser. Photos are not sent to a server for inference.
HOW IT WORKSStart with a photo.
Start with a photo.
Continue with insight.
No app installation or account needed. Use a photo from your gallery or device camera.
Choose a leaf photo
Use a clear, focused photo with adequate light.
Run a check
Follow the process illustration while the model prepares results.
Review, then discuss
Compare the prediction with the original photo and the plant's field conditions.
TECHNOLOGYTwo networks.
Two networks.
Complementary roles.
Segmentation maps possible signs. The classifier combines image context and region features to recognize disease.
RICELEAF / MODEL ARCHITECTURE
U-Net++
EfficientNetB0 encoder · Nested skip connections
ENCODEREfficientNetB0
High resolution features
Multi scale features
Deep semantic features
Nested connections · Simplified schematic
DECODER + HEADU-Net++ decoder
Multi scale feature fusion
Logits · 1 × 320 × 320
Sigmoid · Probability map
Segmentation Guided Classifier
EfficientNetB0 → 1.280 × 10 × 10 feature map
Global average pooling
1.280 global features
Probability weighted pooling
1.280 region features
18 shape features
Linear 18 → 32 · ReLU
Concatenate · 2.592 features → Dropout 0,3 → Linear 2.592 → 4 → Softmax
Two EfficientNetB0 backbones with different roles, executed within one ONNX model.
Understand model capabilities and limitations
Recognizes four rice diseases. It does not identify healthy leaves or check whether a photo shows a rice leaf. No highlighted area does not necessarily mean the leaf is healthy. Highlighted regions do not measure disease severity. Review the result alongside the plant’s condition and expert advice.
FROM THE FIELD, FOR DIFFERENT ROLESTechnology closer
Technology closer
to its users.
ABOUT RICELEAFBuilt with curiosity.
Built with curiosity.
Growing with users.
Developed by Wawu Tri Ambodo, RiceLeaf began as an exploration of rice leaf disease recognition. We are preparing product development through user feedback and opportunities for pilot testing.
↗Let's start
Let's start
a conversation.
For farmers, advisers, and organizations interested in exploring a RiceLeaf pilot.