app.py

import gradio as gr
import torch
from transformers import AutoProcessor, AutoModel # Keep CLIP for potential future use or if FastSAM's text prompt isn't enough
from PIL import Image, ImageDraw, ImageFont
import numpy as np
import random
import os
import wget # To download weights
import traceback # For detailed error printing

# --- Configuration & Model Loading ---

# Device Selection
DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
# Force CPU if CUDA fails or isn't desired (sometimes needed on Spaces free tier)
# DEVICE = "cpu"
print(f"Using device: {DEVICE}")

# --- CLIP Setup (Kept in case needed, but FastSAM's method is primary now) ---
CLIP_MODEL_ID = "openai/clip-vit-base-patch32"
clip_processor = None
clip_model = None

def load_clip_model():
    global clip_processor, clip_model
    if clip_processor is None:
        try:
            print(f"Loading CLIP processor: {CLIP_MODEL_ID}...")
            clip_processor = AutoProcessor.from_pretrained(CLIP_MODEL_ID)
            print("CLIP processor loaded.")
        except Exception as e:
            print(f"Error loading CLIP processor: {e}")
            return False # Indicate failure
    if clip_model is None:
        try:
            print(f"Loading CLIP model: {CLIP_MODEL_ID}...")
            clip_model = AutoModel.from_pretrained(CLIP_MODEL_ID).to(DEVICE)
            print(f"CLIP model loaded to {DEVICE}.")
        except Exception as e:
            print(f"Error loading CLIP model: {e}")
            return False # Indicate failure
    return True # Indicate success


# --- FastSAM Setup ---
FASTSAM_CHECKPOINT = "FastSAM-s.pt"
FASTSAM_CHECKPOINT_URL = f"https://huggingface.co/CASIA-IVA-Lab/FastSAM-s/resolve/main/{FASTSAM_CHECKPOINT}"

fastsam_model = None
fastsam_lib_imported = False # Flag to check if import worked

def check_and_import_fastsam():
    global fastsam_lib_imported
    if not fastsam_lib_imported:
        try:
            from fastsam import FastSAM, FastSAMPrompt
            globals()['FastSAM'] = FastSAM # Make classes available globally
            globals()['FastSAMPrompt'] = FastSAMPrompt
            fastsam_lib_imported = True
            print("fastsam library imported successfully.")
        except ImportError:
            print("Error: 'fastsam' library not found or import failed.")
            print("Please ensure 'fastsam' is installed correctly (pip install fastsam).")
            fastsam_lib_imported = False
        except Exception as e:
            print(f"An unexpected error occurred during fastsam import: {e}")
            traceback.print_exc()
            fastsam_lib_imported = False
    return fastsam_lib_imported


def download_fastsam_weights():
    if not os.path.exists(FASTSAM_CHECKPOINT):
        print(f"Downloading FastSAM weights: {FASTSAM_CHECKPOINT} from {FASTSAM_CHECKPOINT_URL}...")
        try:
            wget.download(FASTSAM_CHECKPOINT_URL, FASTSAM_CHECKPOINT)
            print("FastSAM weights downloaded.")
        except Exception as e:
            print(f"Error downloading FastSAM weights: {e}")
            print("Please ensure the URL is correct and reachable, or manually place the weights file.")
            if os.path.exists(FASTSAM_CHECKPOINT):
                 try: os.remove(FASTSAM_CHECKPOINT)
                 except OSError: pass
            return False
    return os.path.exists(FASTSAM_CHECKPOINT)

def load_fastsam_model():
    global fastsam_model
    if fastsam_model is None:
        if not check_and_import_fastsam():
             print("Cannot load FastSAM model because the library couldn't be imported.")
             return False # Indicate failure

        if download_fastsam_weights():
            try:
                print(f"Loading FastSAM model: {FASTSAM_CHECKPOINT}...")
                fastsam_model = FastSAM(FASTSAM_CHECKPOINT)
                # The FastSAM model itself doesn't need explicit .to(DEVICE)
                # It seems to handle device selection internally or via the prompt process
                print(f"FastSAM model loaded.")
                return True # Indicate success
            except Exception as e:
                print(f"Error loading FastSAM model: {e}")
                traceback.print_exc()
        else:
            print("FastSAM weights not found or download failed. Cannot load model.")
    return fastsam_model is not None # Return True if already loaded or loaded successfully


# --- Processing Functions ---

# (Keep run_clip_zero_shot and run_fastsam_segmentation as they were for the other tabs)
# CLIP Zero-Shot Classification Function
def run_clip_zero_shot(image: Image.Image, text_labels: str):
    # Load CLIP if needed
    if clip_model is None or clip_processor is None:
        if not load_clip_model():
            return "Error: CLIP Model could not be loaded. Check logs.", None

    if image is None: return "Please upload an image.", None
    if not text_labels: return {}, image # Return empty dict, show image

    labels = [label.strip() for label in text_labels.split(',') if label.strip()]
    if not labels: return {}, image

    print(f"Running CLIP zero-shot classification with labels: {labels}")
    try:
        if image.mode != "RGB": image = image.convert("RGB")
        inputs = clip_processor(text=labels, images=image, return_tensors="pt", padding=True).to(DEVICE)
        with torch.no_grad():
            outputs = clip_model(**inputs)
            probs = outputs.logits_per_image.softmax(dim=1)
        print("CLIP processing complete.")
        confidences = {labels[i]: float(probs[0, i].item()) for i in range(len(labels))}
        return confidences, image
    except Exception as e:
        print(f"Error during CLIP processing: {e}")
        traceback.print_exc()
        return f"An error occurred during CLIP: {e}", image

# FastSAM Everything Segmentation Function (for the second tab)
def run_fastsam_segmentation(image_pil: Image.Image, conf_threshold: float = 0.4, iou_threshold: float = 0.9):
    if not load_fastsam_model():
        return "Error: FastSAM Model not loaded. Check logs."
    if not fastsam_lib_imported:
        return "Error: FastSAM library not available."
    if image_pil is None: return "Please upload an image."

    print("Running FastSAM 'segment everything'...")
    try:
        if image_pil.mode != "RGB": image_pil = image_pil.convert("RGB")
        image_np_rgb = np.array(image_pil)

        everything_results = fastsam_model(
            image_np_rgb, device=DEVICE, retina_masks=True, imgsz=640,
            conf=conf_threshold, iou=iou_threshold,
        )
        prompt_process = FastSAMPrompt(image_np_rgb, everything_results, device=DEVICE)
        ann = prompt_process.everything_prompt()
        print(f"FastSAM 'everything' found {len(ann[0]['masks']) if ann and ann[0] and 'masks' in ann[0] else 0} masks.")

        # Plotting
        output_image = image_pil.copy()
        if ann and ann[0] is not None and 'masks' in ann[0] and len(ann[0]['masks']) > 0:
            masks = ann[0]['masks'].cpu().numpy()
            overlay = Image.new('RGBA', output_image.size, (0, 0, 0, 0))
            draw = ImageDraw.Draw(overlay)
            for mask in masks:
                color = (random.randint(50, 255), random.randint(50, 255), random.randint(50, 255), 128)
                mask_image = Image.fromarray((mask * 255).astype(np.uint8), mode='L')
                draw.bitmap((0, 0), mask_image, fill=color)
            output_image = Image.alpha_composite(output_image.convert('RGBA'), overlay).convert('RGB')

        print("FastSAM 'everything' processing complete.")
        return output_image

    except Exception as e:
        print(f"Error during FastSAM 'everything' processing: {e}")
        traceback.print_exc()
        return f"An error occurred during FastSAM 'everything': {e}"


# --- NEW: Text-Prompted Segmentation Function ---
def run_text_prompted_segmentation(image_pil: Image.Image, text_prompts: str, conf_threshold: float = 0.4, iou_threshold: float = 0.9):
    """Segments objects based on text prompts."""
    if not load_fastsam_model():
        return "Error: FastSAM Model not loaded. Check logs.", "No prompts provided."
    if not fastsam_lib_imported:
        return "Error: FastSAM library not available.", "FastSAM library error."
    if image_pil is None:
        return "Please upload an image.", "No image provided."
    if not text_prompts:
        return image_pil, "Please enter text prompts (e.g., 'person, dog')." # Return original image and message

    prompts = [p.strip() for p in text_prompts.split(',') if p.strip()]
    if not prompts:
        return image_pil, "No valid text prompts entered."

    print(f"Running FastSAM text-prompted segmentation for: {prompts}")

    try:
        if image_pil.mode != "RGB":
            image_pil = image_pil.convert("RGB")
        image_np_rgb = np.array(image_pil)

        # 1. Run FastSAM once to get all potential results
        # NOTE: We might optimize later, but this is the standard way FastSAMPrompt works.
        everything_results = fastsam_model(
            image_np_rgb, device=DEVICE, retina_masks=True, imgsz=640,
            conf=conf_threshold, iou=iou_threshold, verbose=False # Less console spam
        )

        # 2. Create the prompt processor
        prompt_process = FastSAMPrompt(image_np_rgb, everything_results, device=DEVICE)

        # 3. Use text_prompt for each prompt and collect masks
        all_matching_masks = []
        found_prompts = []

        for text in prompts:
            print(f"  Processing prompt: '{text}'")
            # Ann is a list of dictionaries, one per image. We have one image.
            # Each dict can have 'masks', 'bboxes', 'points'.
            # text_prompt filters 'everything_results' based on CLIP-like similarity.
            # It might return multiple masks if multiple instances match the text.
            ann = prompt_process.text_prompt(text=text)

            if ann and ann[0] is not None and 'masks' in ann[0] and len(ann[0]['masks']) > 0:
                num_found = len(ann[0]['masks'])
                print(f"    Found {num_found} mask(s) matching '{text}'.")
                found_prompts.append(f"{text} ({num_found})")
                masks = ann[0]['masks'].cpu().numpy() # Get masks as numpy array (N, H, W)
                all_matching_masks.extend(masks) # Add the numpy arrays to the list
            else:
                print(f"    No masks found matching '{text}'.")
                found_prompts.append(f"{text} (0)")

        # 4. Plot the collected masks
        output_image = image_pil.copy()
        status_message = f"Found segments for: {', '.join(found_prompts)}" if found_prompts else "No matching segments found for any prompt."

        if not all_matching_masks:
            print("No matching masks found for any prompt.")
            return output_image, status_message # Return original image if nothing matched

        # Convert list of (H, W) masks to a single (N, H, W) array for consistent processing
        masks_np = np.stack(all_matching_masks, axis=0) # Shape (TotalMasks, H, W)

        overlay = Image.new('RGBA', output_image.size, (0, 0, 0, 0))
        draw = ImageDraw.Draw(overlay)

        for i in range(masks_np.shape[0]):
            mask = masks_np[i] # Shape (H, W), boolean
            color = (random.randint(50, 255), random.randint(50, 255), random.randint(50, 255), 150) # RGBA with slightly more alpha
            mask_image = Image.fromarray((mask * 255).astype(np.uint8), mode='L')
            draw.bitmap((0, 0), mask_image, fill=color)

        output_image = Image.alpha_composite(output_image.convert('RGBA'), overlay).convert('RGB')

        print("FastSAM text-prompted processing complete.")
        return output_image, status_message

    except Exception as e:
        print(f"Error during FastSAM text-prompted processing: {e}")
        traceback.print_exc()
        return f"An error occurred: {e}", "Error during processing."


# --- Gradio Interface ---

print("Attempting to preload models...")
# load_clip_model() # Load CLIP lazily if needed
load_fastsam_model() # Load FastSAM eagerly
print("Preloading finished (or attempted).")


with gr.Blocks(theme=gr.themes.Soft()) as demo:
    gr.Markdown("# CLIP & FastSAM Demo")
    gr.Markdown("Explore Zero-Shot Classification, 'Segment Everything', and Text-Prompted Segmentation.")

    with gr.Tabs():
        # --- CLIP Tab (No changes) ---
        with gr.TabItem("CLIP Zero-Shot Classification"):
            # ... (keep the existing layout and logic for CLIP) ...
            gr.Markdown("Upload an image and provide comma-separated candidate labels (e.g., 'cat, dog, car'). CLIP will predict the probability of the image matching each label.")
            with gr.Row():
                with gr.Column(scale=1):
                    clip_input_image = gr.Image(type="pil", label="Input Image")
                    clip_text_labels = gr.Textbox(label="Comma-Separated Labels", placeholder="e.g., astronaut, moon, dog playing fetch")
                    clip_button = gr.Button("Run CLIP Classification", variant="primary")
                with gr.Column(scale=1):
                    clip_output_label = gr.Label(label="Classification Probabilities")
                    clip_output_image_display = gr.Image(type="pil", label="Input Image Preview")
            clip_button.click(
                run_clip_zero_shot,
                inputs=[clip_input_image, clip_text_labels],
                outputs=[clip_output_label, clip_output_image_display]
            )
            gr.Examples(
                examples=[
                    ["examples/astronaut.jpg", "astronaut, moon, rover, mountain"],
                    ["examples/dog_bike.jpg", "dog, bicycle, person, park, grass"],
                    ["examples/clip_logo.png", "logo, text, graphics, abstract art"],
                ],
                inputs=[clip_input_image, clip_text_labels],
                outputs=[clip_output_label, clip_output_image_display], fn=run_clip_zero_shot, cache_examples=False,
            )


        # --- FastSAM Everything Tab (No changes) ---
        with gr.TabItem("FastSAM Segment Everything"):
            # ... (keep the existing layout and logic for segment everything) ...
             gr.Markdown("Upload an image. FastSAM will attempt to segment all objects/regions in the image.")
             with gr.Row():
                 with gr.Column(scale=1):
                     fastsam_input_image_all = gr.Image(type="pil", label="Input Image", elem_id="fastsam_input_all") # Unique elem_id if needed
                     with gr.Row():
                         fastsam_conf_all = gr.Slider(minimum=0.1, maximum=1.0, value=0.4, step=0.05, label="Confidence Threshold")
                         fastsam_iou_all = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.05, label="IoU Threshold")
                     fastsam_button_all = gr.Button("Run FastSAM Segmentation", variant="primary")
                 with gr.Column(scale=1):
                     fastsam_output_image_all = gr.Image(type="pil", label="Segmented Image", elem_id="fastsam_output_all")
             fastsam_button_all.click(
                 run_fastsam_segmentation,
                 inputs=[fastsam_input_image_all, fastsam_conf_all, fastsam_iou_all],
                 outputs=[fastsam_output_image_all]
             )
             gr.Examples(
                 examples=[
                     ["examples/dogs.jpg", 0.4, 0.9],
                     ["examples/fruits.jpg", 0.5, 0.8],
                     ["examples/lion.jpg", 0.45, 0.9],
                 ],
                 inputs=[fastsam_input_image_all, fastsam_conf_all, fastsam_iou_all],
                 outputs=[fastsam_output_image_all], fn=run_fastsam_segmentation, cache_examples=False,
             )

        # --- NEW: Text-Prompted Segmentation Tab ---
        with gr.TabItem("Text-Prompted Segmentation"):
            gr.Markdown("Upload an image and provide comma-separated text prompts (e.g., 'person, dog, backpack'). FastSAM + CLIP (internally) will segment only the objects matching the text.")
            with gr.Row():
                with gr.Column(scale=1):
                    prompt_input_image = gr.Image(type="pil", label="Input Image")
                    prompt_text_input = gr.Textbox(label="Comma-Separated Text Prompts", placeholder="e.g., glasses, watch, t-shirt")
                    with gr.Row(): # Reuse confidence/IoU sliders if desired
                        prompt_conf = gr.Slider(minimum=0.1, maximum=1.0, value=0.4, step=0.05, label="Confidence Threshold")
                        prompt_iou = gr.Slider(minimum=0.1, maximum=1.0, value=0.9, step=0.05, label="IoU Threshold")
                    prompt_button = gr.Button("Segment by Text", variant="primary")
                with gr.Column(scale=1):
                    prompt_output_image = gr.Image(type="pil", label="Text-Prompted Segmentation")
                    prompt_status_message = gr.Textbox(label="Status", interactive=False) # To show which prompts matched

            prompt_button.click(
                run_text_prompted_segmentation,
                inputs=[prompt_input_image, prompt_text_input, prompt_conf, prompt_iou],
                outputs=[prompt_output_image, prompt_status_message] # Map to image and status box
            )
            gr.Examples(
                examples=[
                    ["examples/dog_bike.jpg", "person, bicycle", 0.4, 0.9],
                    ["examples/astronaut.jpg", "person, helmet", 0.35, 0.9],
                    ["examples/dogs.jpg", "dog", 0.4, 0.9], # Should find multiple dogs
                    ["examples/fruits.jpg", "banana, apple", 0.5, 0.8],
                    ["examples/teacher.jpg", "person, glasses, blackboard", 0.4, 0.9], # Download this image or use another one with glasses/blackboard
                ],
                inputs=[prompt_input_image, prompt_text_input, prompt_conf, prompt_iou],
                outputs=[prompt_output_image, prompt_status_message],
                fn=run_text_prompted_segmentation,
                cache_examples=False,
            )

    # Ensure example images exist or are downloaded
    # (Keep the existing example download logic, maybe add teacher.jpg if used in examples)
    if not os.path.exists("examples"):
        os.makedirs("examples")
        print("Created 'examples' directory. Attempting to download sample images...")
        example_files = {
             "astronaut.jpg": "https://upload.wikimedia.org/wikipedia/commons/thumb/d/d1/Astronaut_-_St._Jean_Bay.jpg/640px-Astronaut_-_St._Jean_Bay.jpg",
             "dog_bike.jpg": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/gradio/outputs_multimodal.jpg",
             "clip_logo.png": "https://raw.githubusercontent.com/openai/CLIP/main/CLIP.png",
             "dogs.jpg": "https://raw.githubusercontent.com/ultralytics/assets/main/im/image8.jpg",
             "fruits.jpg": "https://raw.githubusercontent.com/ultralytics/assets/main/im/image9.jpg",
             "lion.jpg": "https://huggingface.co/spaces/gradio/image-segmentation/resolve/main/images/lion.jpg",
             "teacher.jpg": "https://images.pexels.com/photos/848117/pexels-photo-848117.jpeg?auto=compress&cs=tinysrgb&w=600" # Example with glasses/board
        }
        for filename, url in example_files.items():
             filepath = os.path.join("examples", filename)
             if not os.path.exists(filepath):
                 try:
                     print(f"Downloading {filename}...")
                     wget.download(url, filepath)
                 except Exception as e:
                     print(f"Could not download {filename} from {url}: {e}")
        print("Example image download attempt finished.")


# Launch the Gradio app
if __name__ == "__main__":
    demo.launch(debug=True) # debug=True is helpful locally