Kenyan-Food-13 Classifier

This is a ResNet-18 model fine-tuned on 13 Kenyan dishes.

Model Details

This repository contains a fine-tuned ResNet-18 image classifier for 13 Kenyan dishes:

• Architecture: ResNet-18 (pretrained on ImageNet, final layer re-trained)

• Labels (id2label):

  
  0 → bhaji
  1 → chapati
  2 → githeri
  3 → kachumbari
  4 → kukuchoma
  5 → mandazi
  6 → masalachips
  7 → matoke
  8 → mukimo
  9 → nyamachoma
  10 → pilau
  11 → sukumawiki
  12 → ugali
  

It was trained on a custom “Food13” folder structure.
Model architecture: ResNet-18 (pretrained on ImageNet, then last layer re-trained).

Files

• pytorch_model.bin — PyTorch state dict of the trained ResNet-18.
• config.json — Hugging Face configuration describing number of labels and id2label mapping.
• README.md — (this file).

Usage

from transformers import AutoImageProcessor, AutoModelForImageClassification
from PIL import Image

MODEL_ID = "walter-taya/kenyan-food-13"

# 1. Load a compatible image processor + model:
processor = AutoImageProcessor.from_pretrained(MODEL_ID)
model = AutoModelForImageClassification.from_pretrained(MODEL_ID)

# 2. Preprocess & predict:
image = Image.open("some_kenyan_dish.jpg").convert("RGB")
inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)
logits = outputs.logits
probs = logits.softmax(dim=1)
predicted_class_idx = probs.argmax(-1).item()
label = model.config.id2label[str(predicted_class_idx)]
print(f"Predicted dish: {label} (score={probs[0,predicted_class_idx]:.3f})")

---

license: apache-2.0