import gradio as gr
from transformers import SegformerForSemanticSegmentation, SegformerImageProcessor
from PIL import Image
import torch
import numpy as np

# Load pretrained model
processor = SegformerImageProcessor(do_reduce_labels=False)
model = SegformerForSemanticSegmentation.from_pretrained("nvidia/segformer-b0-finetuned-ade-512-512")
model.eval()

# Prediction function
def segment_image(input_image):
    inputs = processor(images=input_image, return_tensors="pt")

    with torch.no_grad():
        outputs = model(**inputs)
        logits = outputs.logits

    pred_mask = torch.argmax(logits, dim=1)[0].cpu().numpy()
    normalized_mask = (pred_mask * (255 // logits.shape[1])).astype(np.uint8)
    output_image = Image.fromarray(normalized_mask)

    # Bigger mask (3x)
    scale_factor = 3
    new_size = (output_image.width * scale_factor, output_image.height * scale_factor)
    bigger_output = output_image.resize(new_size, resample=Image.NEAREST)

    return bigger_output

# Gradio Interface with submit button (live=False)
demo = gr.Interface(
    fn=segment_image,
    inputs=gr.Image(type="pil", label="Upload Blood Smear Image"),
    outputs=gr.Image(type="pil", label="Predicted Grayscale Mask"),
    title="Malaria Blood Smear Segmentation (SegFormer - Pretrained)",
    description="Upload a blood smear image to segment it using a pretrained SegFormer model (ADE20K 150 classes).",
    examples=["1.png", "3.png"],
    live=False  # <<< ye laga dena
)

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