ChashiBhAI on-device disease classifiers
Collection
YOLO26-cls TFLite classifiers for ChashiBhAI. Rice, brassica, and corn independently tested. • 6 items • Updated • 1
ChashiBhAI On-Device Disease Classifiers (suite index)
Author: Shaq2 (Shakil Ahmed)
Collection: ChashiBhAI on-device disease classifiers
This repository is the suite landing page. Each crop is published as its own model (Hub convention: one model card, one artifact set, independent versioning).
| Model | Classes | TFLite | Preprocess vs .pt |
Role in app |
|---|---|---|---|---|
| chashibhai-rice-disease-cls | 8 | ~3.1 MB | Verified (variant C, 100% top-1) | Bundled in APK — demo primary |
| chashibhai-brassica-disease-cls | 11 | ~10.9 MB | Verified | Model Manager download |
| chashibhai-corn-disease-cls | 4 | ~10.9 MB | Verified | Model Manager download |
This folder still hosts a combined snapshot under tflite/ and pytorch/ for a single-click download of the whole pack. Prefer the per-crop repos for citation and app wiring.
Validation
Independent test on a separate machine (August 2026) confirmed all three TFLite exports load and classify correctly with variant-C preprocessing. Corn is no longer a research preview.
Contract (all three)
- YOLO26-cls,
nms: false, softmaxed[1, nClasses] - Preprocess variant C: shortest-side resize → centre-crop →
/255.0RGB NHWC labels.jsonis canonical (never hardcode class names or preprocess)
What this is not
KrishokChat LLM, RAG, and extra .pt packs belong to RaiyanKhaan/KrishokChat-Advisory-System (arXiv:2606.29243). Those are cited, not claimed.
App
ChashiBhAI — Expo React Native, Bangla-first, on-device diagnosis.
Code: GitHub
License
MIT. See LICENSE and NOTICE.md.
- Downloads last month
- 67