app.py

import gradio as gr
import torch
torch.jit.script = lambda f: f  # Avoid script error in lambda
from t2v_metrics import VQAScore, list_all_vqascore_models

def update_model(model_name):
    return VQAScore(model=model_name, device="cuda")

# Use global variables for model pipe and current model name
global model_pipe, cur_model_name
cur_model_name = "clip-flant5-xl"
model_pipe = update_model(cur_model_name)

# Ensure GPU context manager is imported correctly (assuming spaces is a module you have)
try:
    from spaces import GPU
except ImportError:
    GPU = lambda duration: (lambda f: f)  # Dummy decorator if spaces.GPU is not available

@GPU(duration=20)
def generate(model_name, image, text):
    global model_pipe, cur_model_name
    
    if model_name != cur_model_name:
        cur_model_name = model_name  # Update the current model name
        model_pipe = update_model(model_name)
    
    print("Image:", image)  # Debug: Print image path
    print("Text:", text)  # Debug: Print text input
    print("Using model:", model_name)
    
    try:
        result = model_pipe(images=[image], texts=[text]).cpu()[0][0].item()  # Perform the model inference
        print("Result:", result)
    except RuntimeError as e:
        print(f"RuntimeError during model inference: {e}")
        raise e
    
    return result

@GPU(duration=20)
def rank_images(model_name, images, text):
    global model_pipe, cur_model_name
    
    if model_name != cur_model_name:
        cur_model_name = model_name  # Update the current model name
        model_pipe = update_model(model_name)

    images = [image_tuple[0] for image_tuple in images]
    print("Images:", images)  # Debug: Print image paths
    print("Text:", text)  # Debug: Print text input
    print("Using model:", model_name)
    
    try:
        results = model_pipe(images=images, texts=[text]).cpu()[:, 0].tolist()  # Perform the model inference on all images
        print("Initial results: should be imgs x texts", results)
        ranked_results = sorted(zip(images, results), key=lambda x: x[1], reverse=True)  # Rank results
        ranked_images = [(img, f"Rank: {rank + 1} - Score: {score:.2f}") for rank, (img, score) in enumerate(ranked_results)]  # Pair images with their scores and rank
        print("Ranked Results:", ranked_results)
    except RuntimeError as e:
        print(f"RuntimeError during model inference: {e}")
        raise e
    
    return ranked_images


### EXAMPLES ###
example_imgs = ["0_imgs/DALLE3.png", 
                "0_imgs/DeepFloyd.jpg",
                "0_imgs/Midjourney.jpg",
                "0_imgs/SDXL.jpg"]
example_prompt0 = "Two dogs of different breeds playfully chasing around a tree"
example_prompt1 = "Two dogs of the same breed playing on the grass"
###

# Create the first demo
demo_vqascore = gr.Interface(
    fn=generate,  # function to call
    inputs=[
        gr.Dropdown(["clip-flant5-xl", "clip-flant5-xxl"], label="Model Name"), 
        gr.Image(type="filepath"), 
        gr.Textbox(label="Prompt")
    ],  # define the types of inputs
    examples=[
        ["clip-flant5-xl", example_imgs[0], example_prompt0],
        ["clip-flant5-xl", example_imgs[0], example_prompt1],
        ],
    outputs="number",  # define the type of output
    title="VQAScore",  # title of the app
    description="This model evaluates the similarity between an image and a text prompt."
)

# Create the second demo
demo_vqascore_ranking = gr.Interface(
    fn=rank_images,  # function to call
    inputs=[
        gr.Dropdown(["clip-flant5-xl", "clip-flant5-xxl"], label="Model Name"), 
        gr.Gallery(label="Generated Images"), 
        gr.Textbox(label="Prompt")
    ],  # define the types of inputs
    outputs=gr.Gallery(label="Ranked Images"),  # define the type of output
    examples=[
        ["clip-flant5-xl", [[img, ""] for img in example_imgs], example_prompt0],
        ["clip-flant5-xl", [[img, ""] for img in example_imgs], example_prompt1]
        
    ],
    title="VQAScore Ranking",  # title of the app
    description="This model ranks a gallery of images based on their similarity to a text prompt.",
    allow_flagging='never'  
)

# Custom component for loading examples
def load_example(model_name, images, prompt):
    return model_name, images, prompt

# Create the second demo
with gr.Blocks() as demo_vqascore_ranking_beta:
    gr.Markdown("# VQAScore Ranking\nThis model ranks a gallery of images based on their similarity to a text prompt.")
    model_dropdown = gr.Dropdown(["clip-flant5-xl", "clip-flant5-xxl"], label="Model Name")
    gallery = gr.Gallery(label="Generated Images")
    prompt = gr.Textbox(label="Prompt")
    rank_button = gr.Button("Rank Images")
    ranked_gallery = gr.Gallery(label="Ranked Images with Scores")
    
    rank_button.click(fn=rank_images, inputs=[model_dropdown, gallery, prompt], outputs=ranked_gallery)
    
    # Custom example buttons
    example1_button = gr.Button("Load Example 1")
    example2_button = gr.Button("Load Example 2")
    
    example1_button.click(fn=lambda: load_example("clip-flant5-xl", example_imgs, example_prompt0), inputs=[], outputs=[model_dropdown, gallery, prompt])
    example2_button.click(fn=lambda: load_example("clip-flant5-xl", example_imgs, example_prompt1), inputs=[], outputs=[model_dropdown, gallery, prompt])

    # Layout to allow user to input their own data
    with gr.Row():
        gr.Column([model_dropdown, gallery, prompt, rank_button])
        gr.Column([example1_button, example2_button])
        

# Combine the demos into a tabbed interface
tabbed_interface = gr.TabbedInterface([demo_vqascore, demo_vqascore_ranking, demo_vqascore_ranking_beta], ["VQAScore", "VQAScore Ranking"])

# Launch the tabbed interface
tabbed_interface.queue()
tabbed_interface.launch(share=False)