silversur4/indian-lentils-image-classifier

Indian Lentils Image Classifier

ResNet18, fine-tuned to classify 20 varieties of Indian lentils, beans, and chickpeas from a photo.

Model details

• Base model: resnet18, ImageNet-pretrained (torchvision ResNet18_Weights.DEFAULT), fine-tuned end-to-end.
• Input: 224×224 RGB image, ImageNet mean/std normalization (Resize(256) → CenterCrop(224)).
• Output: 20-way classification over the classes listed below.
• Task: image-classification

Training data

Indian Lentils dataset (Mendeley Data) — 10,000 images across 20 classes, split 80/20 (train/eval), stratified by class.

Classes: Black_Beans, Black_EyedPea, Brown_Chickpea, Desi_chickpea, Horse_Gram, Husked_BlackGram, Husked_RedLentil, Kidney_Bean, Red_Beans, Red_KidneyBean, Red_chickpea, Soya_Bean, White_Chickpea, White_beans, White_peabeans, Whole_BlackGram, Whole_GreenGram, Whole_TurkishGram, chana1, chana2.

Evaluation results

Metric	Value
Accuracy (held-out, 2,000 images)	99.9%
Macro F1	1.00

Residual misclassifications were concentrated among visually similar look-alike classes (e.g. Kidney_Bean / Red_Beans / Red_KidneyBean), consistent with genuine generalization rather than a data-leakage artifact.

A reference benchmark study over 18 CNN architectures on this same dataset concluded EfficientNetB0 was the strongest performer; this model matches that accuracy while training ~1.55x faster per epoch on an NVIDIA T4, via per-architecture learning-rate tuning rather than a new technique. Full methodology and GPU performance analysis (mixed precision, torch.compile, cross-architecture comparison) can be found here (link to be inserted).

How to use

import torch
import torch.nn as nn
from torchvision import models, transforms
from huggingface_hub import hf_hub_download
from PIL import Image

weights_path = hf_hub_download(
    repo_id="silversur4/indian-lentils-image-classifier",
    filename="resnet_lentils.pth",
)

model = models.resnet18(weights=None)
model.fc = nn.Linear(model.fc.in_features, 20)
model.load_state_dict(torch.load(weights_path, map_location="cpu"))
model.eval()

transform = transforms.Compose([
    transforms.Resize(256),
    transforms.CenterCrop(224),
    transforms.ToTensor(),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

image = Image.open("your_photo.jpg").convert("RGB")
x = transform(image).unsqueeze(0)
with torch.no_grad():
    probs = torch.softmax(model(x), dim=1)[0]

License

MIT.