app.py

import os
from huggingface_hub import Repository
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt

# Retrieve the token from the environment variables
token = os.environ.get("token")

repo = Repository(
    local_dir="SVD",
    repo_type="model",
    clone_from="robocan/GeoG_City",
    token=token
)
repo.git_pull()

import torch
from torch.utils.data import Dataset, DataLoader
import pandas as pd
import numpy as np
import io
import joblib
import requests
from tqdm import tqdm
from PIL import Image
from torchvision import transforms
from sklearn.preprocessing import LabelEncoder
from sklearn.model_selection import train_test_split
from torchvision import models
import gradio as gr

device = 'cpu'
le = LabelEncoder()
le = joblib.load("SVD/le.gz")
len_classes = len(le.classes_) + 1

class ModelPre(torch.nn.Module):
    def __init__(self):
        super().__init__()
        self.embedding = torch.nn.Sequential(
            *list(models.convnext_small(weights=models.ConvNeXt_Small_Weights.IMAGENET1K_V1).children())[:-1],
            torch.nn.Flatten(),
            torch.nn.Linear(in_features=768,out_features=512),
            torch.nn.ReLU(),
            torch.nn.Linear(in_features=512,out_features=len_classes),
        )
        # Freeze all layers

    def forward(self, data):
        return self.embedding(data)

model = torch.load("SVD/GeoG.pth", map_location=torch.device(device))

modelm = ModelPre()
modelm.load_state_dict(model['model'])

import warnings
warnings.filterwarnings("ignore", category=RuntimeWarning, module="multiprocessing.popen_fork")

cmp = transforms.Compose([
    transforms.ToTensor(),
    transforms.Resize(size=(224, 224), antialias=True),
    transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225]),
])

def predict(input_img):
    with torch.inference_mode():
        img = cmp(input_img).unsqueeze(0)
        res = modelm(img.to(device))
        probabilities = torch.softmax(res, dim=1).cpu().numpy().flatten()
        top_10_indices = np.argsort(probabilities)[-10:][::-1]
        top_10_probabilities = probabilities[top_10_indices]
        top_10_predictions = le.inverse_transform(top_10_indices)
        
        results = {top_10_predictions[i]: float(top_10_probabilities[i]) for i in range(10)}
        return results

def create_bar_plot(predictions):
    data = pd.DataFrame(list(predictions.items()), columns=["Location", "Probability"])
    max_prob = data["Probability"].max()
    return gr.BarPlot(
        data,
        x="Location",
        y="Probability",
        title="Top 10 Predictions with Probabilities",
        tooltip=["Location", "Probability"],
        y_lim=[0, max_prob],
        width=800,  # Set the width of the plot
        height=600  # Set the height of the plot
    )

def predict_and_plot(input_img):
    predictions = predict(input_img)
    return create_bar_plot(predictions)
    
    
gradio_app = gr.Interface(
    fn=predict_and_plot,
    inputs=gr.Image(label="Upload an Image", type="pil"),
    outputs=gr.BarPlot(),
    title="Predict the Location of this Image"
)

if __name__ == "__main__":
    gradio_app.launch()