app.py

import gradio as gr
from gradio_client import Client
import os

BACKEND_URL = os.environ.get("BACKEND_URL", "").strip()

# Create persistent client
try:
    client = Client(BACKEND_URL, headers={"ngrok-skip-browser-warning": "true"})
    backend_available = True
except:
    client = None
    backend_available = False

def forward_to_backend(fn_name, *args):
    """Generic function to forward any call to backend"""
    if not client:
        return [gr.update() for _ in range(7)]
    try:
        return client.predict(*args, api_name=fn_name)
    except Exception as e:
        print(f"Error calling {fn_name}: {e}")
        return [gr.update() for _ in range(7)]

# Wrapper functions
def handle_file_upload_preview(file_obj):
    return forward_to_backend("/handle_file_upload_preview", file_obj)

def handle_webcam_capture(snapshot):
    return forward_to_backend("/handle_webcam_capture", snapshot)

def process_media(file_obj, webcam_img, model_type, conf_thresh, max_dets, task_type):
    return forward_to_backend("/process_media", file_obj, webcam_img, model_type, conf_thresh, max_dets, task_type)

def clear_all_media_and_outputs():
    return forward_to_backend("/clear_all_media_and_outputs")

def handle_example_select(evt: gr.SelectData):
    """Handle example selection locally and update input_file"""
    # Extract the image path from the dataset
    if isinstance(evt.value, dict) and 'image' in evt.value:
        return evt.value['image']
    return None

# Build the interface
with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("<center><h1>PrimateFace Detection, Pose Estimation, and Gaze Estimation Demo</h1></center>")
    
    if not backend_available:
        gr.Markdown("### 🔴 GPU Server Offline - Please check back later")
    else:
        gr.Markdown("Upload an image/video or use your webcam. For webcam, press 'Enter' to take a snapshot.")
        gr.Markdown("Click 'Detect Faces' for results.")
        
        with gr.Row():
            with gr.Column(scale=1):
                with gr.Tabs():
                    with gr.TabItem("Upload File"):
                        input_file = gr.File(label="Upload Image or Video Here", file_types=["image", ".mp4", ".avi", ".mov", ".mkv", ".webm", ".gif"])
                        display_raw_image_file = gr.Image(label="Raw Image Preview", type="pil", interactive=False, visible=False)
                        display_raw_video_file = gr.Video(label="Raw Video Preview", interactive=False, visible=False)
                    
                    with gr.TabItem("Webcam"):
                        gr.Markdown("**Using the Webcam:** Click on feed or press Enter to capture")
                        input_webcam = gr.Image(sources=["webcam"], type="pil", label="Live Webcam")
                        display_raw_image_webcam = gr.Image(label="Captured Snapshot Preview", type="pil", interactive=False, visible=False)
                
                clear_all_button = gr.Button("Clear All Inputs & Outputs")
            
            with gr.Column(scale=1):
                gr.Markdown("### Processed Output")
                display_processed_image = gr.Image(label="Processed Image", type="pil", interactive=False, visible=False)
                display_processed_video = gr.Video(label="Processed Video", interactive=False, visible=False)
        
        # Example images - host them on HF Space
        example_paths = [
            "images/allocebus_000003.jpeg",
            "images/tarsius_000120.jpeg", 
            "images/nasalis_proboscis-monkey.png",
            "images/macaca_000032.jpeg",
            "images/mandrillus_000011.jpeg",
            "images/pongo_000006.jpeg"
        ]
        
        example_dataset = gr.Dataset(
            components=["image"],
            samples=[[path] for path in example_paths],
            label="Example Images (Click to use)",
            samples_per_page=6
        )
        
        submit_button = gr.Button("Detect Faces", variant="primary", scale=2)
        
        with gr.Column():
            gr.Markdown("### Detection Controls")
            model_choice_radio = gr.Radio(choices=["MMDetection"], value="MMDetection", label="Inferencer", visible=False)
            task_type_dropdown = gr.Dropdown(
                choices=["Face Detection", "Face Pose Estimation", "Gaze Estimation [experimental]"],
                value="Face Detection",
                label="Select Task"
            )
            conf_slider = gr.Slider(minimum=0.05, maximum=0.95, value=0.25, step=0.05, label="Confidence Threshold")
            max_det_slider = gr.Slider(minimum=1, maximum=10, value=3, step=1, label="Max Detections")
        
        # Define outputs
        file_preview_outputs = [display_raw_image_file, display_raw_video_file, input_file, display_processed_image, display_processed_video]
        webcam_outputs = [display_raw_image_webcam, input_webcam, display_processed_image, display_processed_video]
        process_outputs = [display_raw_image_file, display_raw_video_file, display_raw_image_webcam, display_processed_image, display_processed_video]
        clear_outputs = [input_file, input_webcam, display_raw_image_file, display_raw_video_file, display_raw_image_webcam, display_processed_image, display_processed_video]
        
        # Wire events
        input_file.change(handle_file_upload_preview, inputs=[input_file], outputs=file_preview_outputs)
        input_webcam.change(handle_webcam_capture, inputs=[input_webcam], outputs=webcam_outputs)
        
        # Handle example selection
        example_dataset.select(handle_example_select, outputs=[input_file])
        
        submit_button.click(
            process_media,
            inputs=[input_file, display_raw_image_webcam, model_choice_radio, conf_slider, max_det_slider, task_type_dropdown],
            outputs=process_outputs
        )
        
        clear_all_button.click(clear_all_media_and_outputs, outputs=clear_outputs)

demo.launch()