app.py

import gradio as gr
import torch
from transformers import ViTFeatureExtractor, ViTForImageClassification
import json

import google.generativeai as genai
from dotenv import load_dotenv
import os
import tempfile

load_dotenv()

genai.configure(api_key=os.getenv("GEMINI_API_KEY"))
generation_config = {
    "temperature": 1,
    "top_p": 0.95,
    "top_k": 64,
    "max_output_tokens": 8192,
    "response_mime_type": "application/json",
}
gmodel = genai.GenerativeModel(
    model_name="gemini-1.5-flash", generation_config=generation_config
)

# model path
MODEL_PATH = "./"
feature_extractor = ViTFeatureExtractor.from_pretrained(MODEL_PATH)
model = ViTForImageClassification.from_pretrained(MODEL_PATH)
model.eval()

def image_detection(image_file):
    try:
        image_file = genai.upload_file(path=image_file)
        prompt = "You are an AI model designed to identify images (of lung cells) if they are histopathological images or not, answer in 0 (not a histopathological image) or 1 (is a histopathological image), we mainly want to remove the most obviously not histopathological images. Format your response in JSON, including fields for 'is_histopathological'."
        response = gmodel.generate_content([prompt, image_file]).text
        genai.delete_file(image_file.name)
        res = json.loads(response)
        return res["is_histopathological"]
    except Exception as e:
        print(e)
        return "error"

# image
def classify_image(image):
    # convert to RGB
    if image.mode != "RGB":
        image = image.convert("RGB")

    # check if image is histopathological
    with tempfile.NamedTemporaryFile(delete=True, suffix=".png") as temp_image:
        image.save(temp_image.name)
        image_path = temp_image.name

        is_histopathological = image_detection(image_path)
        if is_histopathological != 1:
            return {"error": "Please upload a histopathological image."}

    # preprocess
    inputs = feature_extractor(images=image, return_tensors="pt")

    # get prediction
    with torch.no_grad():
        outputs = model(**inputs)
        probabilities = torch.nn.functional.softmax(outputs.logits, dim=-1)

    # results
    results = {
        model.config.id2label[i]: float(prob)
        for i, prob in enumerate(probabilities[0])
    }

    # Convert results to JSON
    json_results = json.dumps(results)

    return json_results

def launch_gradio():
    interface = gr.Interface(
        fn=classify_image,
        inputs=gr.Image(type="pil", label="Upload Cancer Histopathological Image"),
        outputs=gr.JSON(label="Classification Probabilities"),
        title="Lung and Colon Cancer Image Classifier",
        description="Upload a histopathological image to classify cancer type",
    )

    interface.launch(server_name="0.0.0.0", server_port=7860)

if __name__ == "__main__":
    launch_gradio()