app.py

import os
import torch
import numpy as np
import matplotlib.pyplot as plt
from PIL import Image
import gradio as gr
from transformers import ViTFeatureExtractor
from huggingface_hub import hf_hub_download
import spaces
from torchvision import transforms


HF_TOKEN = os.environ.get("HF_TOKEN")
model = None
feature_extractor = None

device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
VALID_DS_PATH = 'valid_ds.pth'
valid_ds = torch.load(VALID_DS_PATH)




from transformers import ViTModel
from transformers.modeling_outputs import SequenceClassifierOutput
import torch.nn as nn
import torch.nn.functional as F

class ViTForImageClassification(nn.Module):
    def __init__(self, num_labels=3):
        super(ViTForImageClassification, self).__init__()
        self.vit = ViTModel.from_pretrained('google/vit-base-patch16-224-in21k')
        self.dropout = nn.Dropout(0.1)
        self.classifier = nn.Linear(self.vit.config.hidden_size, num_labels)
        self.num_labels = num_labels

    def forward(self, pixel_values, labels):
        outputs = self.vit(pixel_values=pixel_values)
        output = self.dropout(outputs.last_hidden_state[:,0])
        logits = self.classifier(output)

        loss = None
        if labels is not None:
          loss_fct = nn.CrossEntropyLoss()
          loss = loss_fct(logits.view(-1, self.num_labels), labels.view(-1))
        if loss is not None:
          return logits, loss.item()
        else:
          return logits, None

# Load an image from file for inference
def load_image(image_path):
    img = Image.open(image_path)
    img = img.convert("RGB")  # Ensure it's in RGB format
    return img


# Inference function
@spaces.GPU()
def run_inference(image, device, valid_ds):
    # Load image from the Gradio input
    # input_image = Image.fromarray(image.astype('uint8'), 'RGB')

    global model, feature_extractor

    if model is None or feature_extractor is None:

        MODEL_PATH = hf_hub_download(repo_id="limitedonly41/offers_26", 
                                           filename="model_50.pt", 
                                           use_auth_token=HF_TOKEN)
        try:
            model = torch.load(MODEL_PATH)
        except:
            model = torch.load(MODEL_PATH, map_location=torch.device('cpu'))
        model.eval()
        model.to(device)
        # feature_extractor = ViTFeatureExtractor.from_pretrained('google/vit-base-patch16-224-in21k', do_rescale=False)

    transform = transforms.Compose([
        transforms.Resize((224, 224)),  # Resize to the model's input size
        transforms.ToTensor(),
    ])
    
    image = Image.fromarray(image.astype('uint8'), 'RGB')
    input_tensor = transform(image)
    input_tensor = input_tensor.unsqueeze(0)  # Add a batch dimension
    input_tensor = input_tensor.to(device)  # Send to appropriate computing device

    # Disable grad
    with torch.no_grad():
        # Generate prediction
        prediction, _ = model(input_tensor, labels=None)

        # Get the predicted class index
        predicted_class = torch.argmax(prediction, dim=1).item()
        value_predicted = list(valid_ds.class_to_idx.keys())[list(valid_ds.class_to_idx.values()).index(predicted_class)]

        # return f"Predicted Class: {value_predicted}, {predicted_class}"
        return value_predicted


    # # Preprocess the image using the feature extractor
    # inputs = feature_extractor(images=input_image, return_tensors="pt")['pixel_values']
    
    # # Send to the appropriate device (CPU/GPU)
    # inputs = inputs.to(device)
    
    # # Disable gradients during inference
    # with torch.no_grad():
    #     # Generate prediction
    #     prediction, _ = model(inputs, None)

    #     # Predicted class value using argmax
    #     predicted_class = np.argmax(prediction.cpu().numpy())
    #     value_predicted = list(valid_ds.class_to_idx.keys())[list(valid_ds.class_to_idx.values()).index(predicted_class)]
        
    #     # Return the result with the predicted class
    #     return f"Predicted Class: {value_predicted}, {predicted_class}"



# Create a Gradio interface
iface = gr.Interface(
    fn=lambda image: run_inference(image, device, valid_ds),
    inputs=gr.Image(type="numpy"),  # Updated to use gr.Image
    outputs="text",  # Output is text (predicted class)
    title="Image Classification",
    description="Upload an image to get the predicted class using the ViT model."
)
# Launch the Gradio app
iface.launch()