README.md

---
tags:
- model_hub_mixin
- pytorch_model_hub_mixin
- image-classification
- densenet121
language:
- ja
---

This model has been pushed to the Hub using the [PytorchModelHubMixin](https://huggingface.co/docs/huggingface_hub/package_reference/mixins#huggingface_hub.PyTorchModelHubMixin) integration:

# Densenet121-Dog-Emotions Model Card

- **学習データでの正答率:** 0.6451
- **テストデータでの正答率:** 0.5938
  
# モデルについて
このモデルは犬の画像を[angry, happy, relaxed, sad]の4つのカテゴリに分類するモデルです。<br>
densenet121の末端に線形層を追加し、devzohaib/dog-emotions-predictionデータセットで微調整を行ないました。参考文献と使い方は以下のようになっています。


## 使い方
1. モデルの読み込み
```sh
from huggingface_hub import PyTorchModelHubMixin
import torch
import torch.nn as nn
from torchvision import models, transforms
from PIL import Image
import requests

class CustomDenseNet(nn.Module, PyTorchModelHubMixin):
  def __init__(self, class_names):
    super().__init__()
    self.densenet = models.densenet121(pretrained=True)
    num_features = self.densenet.classifier.in_features
    self.densenet.fc = nn.Linear(num_features, len(class_names))

  def forward(self, x):
    outputs = self.densenet(x)
    _, preds = torch.max(outputs, 1)
    probabilities = torch.nn.functional.softmax(outputs, dim=1).squeeze(0)

    predicted_class = class_names[preds.item()]
    predicted_probabilities = {class_names[i]: probabilities[i].item() for i in range(len(class_names))}

    return predicted_class, predicted_probabilities

model_id = "shinyice/densenet121-dog-emotions"
class_names = ['angry', 'happy', 'relaxed', 'sad']
model = CustomDenseNet(class_names)
model = model.from_pretrained(model_id)

device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
```

2. 画像分類
```sh
def dog_emotion(model, url_mode=False, input_image=None):
  img_transforms = transforms.Compose([
      transforms.Resize(224),
      transforms.ToTensor(),
      transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])])
  if url_mode:
    image = Image.open(requests.get(input_image, stream=True).raw).convert('RGB')
  else:
    image = Image.open(input_image).convert('RGB')
  image_tensor = img_transforms(image).unsqueeze(0)
  image_tensor = image_tensor.to(device)

  model.eval()
  with torch.no_grad():
    predicted_class, predicted_probabilities = model(image_tensor)

  return predicted_class, predicted_probabilities, image

url_mode = True
input_image = ""

emotion,probabilities, image = dog_emotion(model=model, url_mode=url_mode, input_image=input_image)
print(emotion,probabilities)
image
```

## 参考文献
- [カスタムデータセットでなるべくかんたんに画像分類器をつくりたい。Pytorchで転移学習](https://qiita.com/john-rocky/items/e386f0aa5232d323db7e)
- [Dog Emotions Prediction](https://www.kaggle.com/datasets/devzohaib/dog-emotions-prediction)
- [Uploading models](https://huggingface.co/docs/hub/models-uploading)