Upload a photo and FloraSense names the species, with confidence scores for the five closest matches — from a model trained on 102 flower categories.
or drag & drop an image here
Point, upload, and get an answer — without handing your photo library to anyone.
Inference runs on a CPU container sized for the job. Most identifications return in under a second once the service is warm.
The image is written to a temporary file, classified, and deleted before the response is sent. Nothing is written to a database or bucket.
Photos are scaled to 1024px and re-encoded on your device before upload, so a 12 MP camera shot never leaves as a 12 MP file.
Five candidates with confidence scores, not one guess — useful when a bloom sits between two similar species.
From tap to species name, every step of the path.
Pick a photo or shoot one from the browser
Scaled to 1024px and re-encoded on device
Sent over HTTPS through a signed CDN origin
EfficientNet-B0 scores all 102 categories
Top five returned, image deleted
Each architecture was fine-tuned on the same 102-category dataset and scored on the same held-out test set.
| Architecture | Top-1 accuracy | Correct |
|---|---|---|
| EfficientNet-B0Live | 86.57% | 709 / 819 |
| DenseNet-121 | 83.15% | 681 / 819 |
| VGG-16 | 77.53% | 635 / 819 |
Measured on 819 held-out test images. EfficientNet-B0 is the model serving this page.
FloraSense is a machine learning demonstration and gets it wrong roughly one time in seven. It recognises only the 102 categories it was trained on, so a flower outside that set will still be matched to something — confidently, and incorrectly.
Never rely on it to decide whether a plant is safe to eat, touch, or give to a person or animal, and do not use it for medical, foraging, agricultural, or conservation decisions. Consult a qualified botanist or your local herbarium instead.