app.py

import torch
import gradio as gr

from utils.imageHandling import hfImageToTensor, preprocessing, postprocessing, loadPreloadedImages, legendHandling
from model.modelLoading import loadModel


## %% CONSTANTS
gta_image_dir = "./preloadedImages/GTAV"
city_image_dir = "./preloadedImages/cityScapes"
turin_image_dir = "./preloadedImages/turin"
device = 'cuda' if torch.cuda.is_available() else 'cpu'


MODELS = {
    # "BISENET-BASE": loadModel('bisenet', device, 'weight_Base.pth'),
    "BISENET-BEST": loadModel('bisenet', device, 'weight_Best.pth'),
    # "BISENETV2-BASE": loadModel('bisenetv2', device, 'weight_Base.pth'),
    "BISENETV2-BEST": loadModel('bisenetv2', device, 'weight_Best.pth')
}

image_list = loadPreloadedImages(gta_image_dir, city_image_dir, turin_image_dir)


# %% prediction on an image

def predict(inputImage: torch.Tensor, model) -> torch.Tensor:
    """
    Predict the segmentation mask for the input image using the provided model.
    
    Args:
        inputImage (torch.Tensor): The input image tensor.
        model (BiSeNet): The BiSeNet model for segmentation.
        
    Returns:
        prediction (torch.Tensor): The predicted segmentation mask.
    """
    with torch.no_grad():
        output = model(preprocessing(inputImage.clone()).to(device))
    output = output[0] if isinstance(output, (tuple, list)) else output
    return output[0].argmax(dim=0, keepdim=True).to(device)  


# %% Gradio interface
def run_prediction(image: gr.Image, selected_model: str)-> tuple[torch.Tensor]:
    if image is None:
        return (gr.update(value=None, visible=False), gr.update(value=f"❌ No image provided for prediction.", visible=True))
    
    if selected_model is None:
        return (gr.update(value=None, visible=False),  gr.update(value=f"❌ No model selected for prediction.", visible=True))
    
    if not isinstance(selected_model, str) or selected_model.strip().upper() not in MODELS:
        return (gr.update(value=None, visible=False),  gr.update(value=f"❌ Invalid model selected.", visible=True))

    try:
        image = hfImageToTensor(image, width=1024, height=512)
        prediction = predict(image, MODELS[selected_model.strip().upper()])
        prediction = postprocessing(prediction)
    except Exception as e:
        return (gr.update(value=None, visible=False),  gr.update(value=f"❌ {str(e)}.", visible=True))
    return (gr.update(value=prediction, visible=True), gr.update(value="", visible=False))

# Gradio UI
with gr.Blocks(title="Semantic Segmentation Predictors") as demo:
    gr.Markdown("# Semantic Segmentation with Real-Time Networks")
    gr.Markdown('A small user interface created to run semantic segmentation on images using Cityscapes-like predictions and real-time segmentation networks.')
    gr.Markdown("Upload an image and choose your preferred model for segmentation, or otherwise use one of the preloaded images.")
    gr.Markdown("The full code for the project is available on [GitHub](https://github.com/Nuzz23/MLDL_SemanticSegmentation).")
    with gr.Row():
        with gr.Column():
            image_input = gr.Image(type="pil", label="Upload image")
            model_selector = gr.Radio(
                choices=[   #"BiSeNet-Base", 
                            "BiSeNet-Best", 
                            # "BiSeNetV2-Base", 
                            "BiSeNetV2-Best"],
                value="BiSeNet-Best",
                label="Select the real time segmentation model"
                )
            submit_btn = gr.Button("Run prediction")
        with gr.Column():
            result_display = gr.Image(label="Model prediction", visible=True)
            error_text = gr.Markdown("", visible=False)
            gr.Markdown("The legend of the classes is the following (format **name** -> **color**)")
            with gr.Row():
                legend = legendHandling()
                for i in range(0, len(legend), 2):
                    with gr.Row():
                        with gr.Column(scale=1):
                            color_box0 = f"""<span style='display:inline-block; width:15px; height:15px; 
                                background-color:rgb({legend[i][3][0]},{legend[i][3][1]},{legend[i][3][2]}); margin-left:6px; border:1px solid #000;'></span>"""
                            gr.HTML(f"<div style='display:flex; align-items:center; margin-bottom:-10px; margin-top:-5px;'><b>{legend[i][1]}</b> → {color_box0}</div>")
                        with gr.Column(scale=1):
                            if i + 1 < len(legend):
                                color_box1 = f"""<span style='display:inline-block; width:15px; height:15px; 
                                    background-color:rgb({legend[i+1][3][0]},{legend[i+1][3][1]},{legend[i+1][3][2]}); margin-left:6px; border:1px solid #000;'></span>"""
                                gr.HTML(f"<div style='display:flex; align-items:center; margin-bottom:-10px; margin-top:-5px;'><b>{legend[i+1][1]}</b> → {color_box1}</div>")
                            else:
                                gr.Markdown("") 


    with gr.Row():
        gr.Markdown("## Preloaded images to be used for testing the model")
        gr.Markdown("""You can use images taken from the Grand Theft Auto V video game, the Cityscapes dataset or 
                    even the city of Turin to be used as input for the model without need for manual upload.""")

    # Mostriamo 4 righe da 5 immagini
    for i in range(0, len(image_list), 5):
        with gr.Row():
            for img in image_list[i:i+5]:
                img_comp = gr.Image(value=img, interactive=False, show_label=False, show_download_button=False, height=180, width=256,
                                    show_fullscreen_button=False, show_share_button=False, mirror_webcam=False)
                img_comp.select(fn=lambda x:x, inputs=img_comp, outputs=image_input)
    
    submit_btn.click(
        fn=run_prediction,
        inputs=[image_input, model_selector],
        outputs=[result_display, error_text],
    )

    gr.Markdown("Made by group 21 semantic segmentation project at Politecnico di Torino 2024/2025")

demo.launch()