dog-breed-resnet50

🐶 Dog Breed Classification – ResNet50

A high-accuracy dog breed classifier trained on 120 breeds

📌 Overview

This repository contains a fine-tuned ResNet50 model for dog breed classification. Given an input dog image, the model predicts the most likely breed among 120 classes.

This model is designed for:

🐕 Dog breed classification

🖼️ Image understanding & retrieval

🧪 Educational deep-learning projects

🚀 Plug-and-play inference in Streamlit / Python / FastAPI

🧠 Model Architecture

Backbone: ResNet50 (ImageNet pretrained)

Head: Fully-connected layer with softmax

Loss: Cross Entropy

Optimizer: AdamW

Training Epochs: 15

Input Size: 224×224

Number of classes: {len(id2breed)}

📁 Files Included File Description resnet50_dog.pth Trained model weights id2breed.json Mapping {class_index → breed_name} README.md Model card (this file)

The mapping file is essential for converting numerical model outputs into human-readable breed names.

🔍 Inference Example (Python) import torch from torchvision import transforms from PIL import Image from huggingface_hub import hf_hub_download import json

1. Download files

repo_id = "djhua0103/dog-breed-resnet50"

ckpt_path = hf_hub_download(repo_id, "resnet50_dog_best.pth") json_path = hf_hub_download(repo_id, "id2breed.json")

2. Load label mapping

with open(json_path, "r") as f: id2breed = json.load(f)

3. Build model

from torchvision.models import resnet50 model = resnet50(weights=None) model.fc = torch.nn.Linear(model.fc.in_features, len(id2breed)) model.load_state_dict(torch.load(ckpt_path, map_location="cpu")) model.eval()

4. Preprocess image

transform = transforms.Compose([ transforms.Resize((224,224)), transforms.ToTensor() ])

img = Image.open("dog.jpg") x = transform(img).unsqueeze(0)

5. Predict

with torch.no_grad(): logits = model(x) prob = logits.softmax(dim=1) idx = prob.argmax().item() breed = id2breed[str(idx)]

print("Predicted breed:", breed)

🚀 Use in Streamlit

This model can be easily integrated into any Streamlit app:

from huggingface_hub import hf_hub_download from multimodal_dog.models.classifier import DogClassifier

model = DogClassifier.from_hf(device="cpu") breed, confidence = model.predict(image_tensor)

📊 Dataset

The dataset is based on the Kaggle Dogs Dataset (https://www.kaggle.com/competitions/dog-breed-identification/), containing images of 120 dog breeds. Labels are derived from labels.csv and converted into class indices using alphabetical ordering.

🏷️ Labels

All breed names are stored in id2breed.json.

📘 License

Specify your license (MIT / Apache / custom).

🙌 Acknowledgements

ResNet50 from PyTorch

Hugging Face Hub

Dataset provider (Kaggle / Stanford Dogs)