app.py

# Author: Huzheng Yang
# %%
import copy
from functools import partial
from io import BytesIO
import os

from einops import rearrange
from matplotlib import pyplot as plt
import matplotlib
USE_HUGGINGFACE_ZEROGPU = os.getenv("USE_HUGGINGFACE_ZEROGPU", "False").lower() in ["true", "1", "yes"]
DOWNLOAD_ALL_MODELS_DATASETS = os.getenv("DOWNLOAD_ALL_MODELS_DATASETS", "False").lower() in ["true", "1", "yes"]

if USE_HUGGINGFACE_ZEROGPU:  # huggingface ZeroGPU, dynamic GPU allocation 
    try:
        import spaces
    except:
        USE_HUGGINGFACE_ZEROGPU = False
        
if USE_HUGGINGFACE_ZEROGPU:
    BATCH_SIZE = 1
else:  # run on local machine
    BATCH_SIZE = 1

import gradio as gr

import torch
import torch.nn.functional as F
from PIL import Image
import numpy as np
import time
import threading

from ncut_pytorch.backbone import extract_features, load_model
from ncut_pytorch.backbone import MODEL_DICT, LAYER_DICT, RES_DICT
from ncut_pytorch import NCUT
from ncut_pytorch import eigenvector_to_rgb, rotate_rgb_cube

DATASET_TUPS = [
    # (name, num_classes)
    ('UCSC-VLAA/Recap-COCO-30K', None),
    ('nateraw/pascal-voc-2012', None),
    ('johnowhitaker/imagenette2-320', 10),
    ('jainr3/diffusiondb-pixelart', None),
    ('nielsr/CelebA-faces', None),
    ('JapanDegitalMaterial/Places_in_Japan', None),
    ('Borismile/Anime-dataset', None),
    ('Multimodal-Fatima/CUB_train', 200),
    ('mrm8488/ImageNet1K-val', 1000),
    ("trashsock/hands-images", 8),
]
DATASET_NAMES = [tup[0] for tup in DATASET_TUPS]
DATASET_CLASSES = [tup[1] for tup in DATASET_TUPS]

from datasets import load_dataset

def download_all_datasets():
    for name in DATASET_NAMES:
        print(f"Downloading {name}")
        try:
            load_dataset(name, trust_remote_code=True)
        except Exception as e:
            print(f"Error downloading {name}: {e}")

def compute_ncut(
    features,
    num_eig=100,
    num_sample_ncut=10000,
    affinity_focal_gamma=0.3,
    knn_ncut=10,
    knn_tsne=10,
    embedding_method="UMAP",
    embedding_metric='euclidean',
    num_sample_tsne=300,
    perplexity=150,
    n_neighbors=150,
    min_dist=0.1,
    sampling_method="fps",
    metric="cosine",
):        
    logging_str = ""
    
    num_nodes = np.prod(features.shape[:-1])
    if num_nodes / 2 < num_eig:
        # raise gr.Error("Number of eigenvectors should be less than half the number of nodes.")
        gr.Warning("Number of eigenvectors should be less than half the number of nodes.\n" f"Setting num_eig to {num_nodes // 2 - 1}.")
        num_eig = num_nodes // 2 - 1
        logging_str += f"Number of eigenvectors should be less than half the number of nodes.\n" f"Setting num_eig to {num_nodes // 2 - 1}.\n"
    
    start = time.time()
    eigvecs, eigvals = NCUT(
        num_eig=num_eig,
        num_sample=num_sample_ncut,
        device="cuda" if torch.cuda.is_available() else "cpu",
        affinity_focal_gamma=affinity_focal_gamma,
        knn=knn_ncut,
        sample_method=sampling_method,
        distance=metric,
        normalize_features=False,
    ).fit_transform(features.reshape(-1, features.shape[-1]))
    # print(f"NCUT time: {time.time() - start:.2f}s")
    logging_str += f"NCUT time: {time.time() - start:.2f}s\n"
    
    start = time.time()
    _, rgb = eigenvector_to_rgb(
        eigvecs,
        method=embedding_method,
        metric=embedding_metric,
        num_sample=num_sample_tsne,
        perplexity=perplexity,
        n_neighbors=n_neighbors,
        min_distance=min_dist,
        knn=knn_tsne,
        device="cuda" if torch.cuda.is_available() else "cpu",
    )
    logging_str += f"{embedding_method} time: {time.time() - start:.2f}s\n"

    rgb = rgb.reshape(features.shape[:-1] + (3,))
    return rgb, logging_str, eigvecs


def dont_use_too_much_green(image_rgb):
    # make sure the foval 40% of the image is red leading
    x1, x2 = int(image_rgb.shape[1] * 0.3), int(image_rgb.shape[1] * 0.7)
    y1, y2 = int(image_rgb.shape[2] * 0.3), int(image_rgb.shape[2] * 0.7)
    sum_values = image_rgb[:, x1:x2, y1:y2].mean((0, 1, 2))
    sorted_indices = sum_values.argsort(descending=True)
    image_rgb = image_rgb[:, :, :, sorted_indices]
    return image_rgb


def to_pil_images(images, target_size=512, resize=True):
    size = images[0].shape[1]
    multiplier = target_size // size
    res = int(size * multiplier)
    pil_images = [
            Image.fromarray((image * 255).cpu().numpy().astype(np.uint8))
            for image in images
        ]
    if resize:
        pil_images = [
            image.resize((res, res), Image.Resampling.NEAREST)
            for image in pil_images
        ]
    return pil_images
    


def pil_images_to_video(images, output_path, fps=5):
    # from pil images to numpy
    images = [np.array(image) for image in images]
    
    # print("Saving video to", output_path)
    import cv2
    fourcc = cv2.VideoWriter_fourcc(*'mp4v')
    height, width, _ = images[0].shape
    out = cv2.VideoWriter(output_path, fourcc, fps, (width, height))
    for image in images:
        out.write(cv2.cvtColor(image, cv2.COLOR_RGB2BGR))
    out.release()
    return output_path

# save up to 100 videos in disk
class VideoCache:
    def __init__(self, max_videos=100):
        self.max_videos = max_videos
        self.videos = {}
    
    def add_video(self, video_path):
        if len(self.videos) >= self.max_videos:
            pop_path = self.videos.popitem()[0]
            try:
                os.remove(pop_path)
            except:
                pass
        self.videos[video_path] = video_path
    
    def get_video(self, video_path):
        return self.videos.get(video_path, None)

video_cache = VideoCache()
    
def get_random_path(length=10):
    import random
    import string
    name = ''.join(random.choices(string.ascii_lowercase + string.digits, k=length))
    path = f'/tmp/{name}.mp4'
    return path

default_images = ['./images/image_0.jpg', './images/image_1.jpg', './images/image_2.jpg', './images/image_3.jpg', './images/guitar_ego.jpg', './images/image_5.jpg']
default_outputs = ['./images/image-1.webp', './images/image-2.webp', './images/image-3.webp', './images/image-4.webp', './images/image-5.webp']
# default_outputs_independent = ['./images/image-6.webp', './images/image-7.webp', './images/image-8.webp', './images/image-9.webp', './images/image-10.webp']
default_outputs_independent = []

downscaled_images = ['./images/image_0_small.jpg', './images/image_1_small.jpg', './images/image_2_small.jpg', './images/image_3_small.jpg', './images/image_5_small.jpg']
downscaled_outputs = default_outputs

example_items = downscaled_images[:3] + downscaled_outputs[:3]


def run_alignedthreemodelattnnodes(images, model, batch_size=16):
    
    use_cuda = torch.cuda.is_available() 
    device = torch.device("cuda" if use_cuda else "cpu")
    
    if use_cuda:
        model = model.to(device)
        
    chunked_idxs = torch.split(torch.arange(images.shape[0]), batch_size)
    
    outputs = []
    for idxs in chunked_idxs:
        inp = images[idxs]
        if use_cuda:
            inp = inp.to(device)
        out = model(inp)  
        # normalize before save
        out = F.normalize(out, dim=-1)
        outputs.append(out.cpu().float())
    outputs = torch.cat(outputs, dim=0)

    return outputs


def _reds_colormap(image):
    # normed_data = image / image.max()  # Normalize to [0, 1]
    normed_data = image
    colormap = matplotlib.colormaps['inferno']  # Get the Reds colormap
    colored_image = colormap(normed_data)  # Apply colormap
    return (colored_image[..., :3] * 255).astype(np.uint8)  # Convert to RGB

# heatmap images
def apply_reds_colormap(images, size):
    # for i_image in range(images.shape[0]):
    #     images[i_image] -= images[i_image].min()
    #     images[i_image] /= images[i_image].max()
    # normed_data = [_reds_colormap(images[i]) for i in range(images.shape[0])]
    # normed_data = np.stack(normed_data)
    normed_data = _reds_colormap(images)
    normed_data = torch.tensor(normed_data).float()
    normed_data = rearrange(normed_data, "b h w c -> b c h w")
    normed_data = torch.nn.functional.interpolate(normed_data, size=size, mode="nearest")
    normed_data = rearrange(normed_data, "b c h w -> b h w c")
    normed_data = normed_data.cpu().numpy().astype(np.uint8)
    return normed_data

# Blend heatmap with the original image
def blend_image_with_heatmap(image, heatmap, opacity1=0.5, opacity2=0.5):
    blended = (1 - opacity1) * image + opacity2 * heatmap
    return blended.astype(np.uint8)

def make_cluster_plot(eigvecs, images, h=64, w=64):
    from ncut_pytorch.ncut_pytorch import farthest_point_sampling
    magnitude = torch.norm(eigvecs, dim=-1)
    p = 0.5
    top_p_idx = magnitude.argsort(descending=True)[:int(p * magnitude.shape[0])]
    num_samples = 50
    fps_idx = farthest_point_sampling(eigvecs[top_p_idx], num_samples)
    fps_idx = top_p_idx[fps_idx]
    
    # downsample to 256x256
    images = F.interpolate(images, (256, 256), mode="bilinear")
    images = images.cpu().numpy()
    images = images.transpose(0, 2, 3, 1)
    images = images * 255
    images = images.astype(np.uint8)
    
    
    # sort the fps_idx by the mean of the heatmap
    fps_heatmaps = {}
    sort_values = []
    for _, idx in enumerate(fps_idx):
        device = 'cuda' if torch.cuda.is_available() else 'cpu'
        eigvecs = eigvecs.to(device)
        heatmap = F.cosine_similarity(eigvecs, eigvecs[idx][None], dim=-1)
        heatmap = heatmap.reshape(-1, h, w)
        mask = (heatmap > 0.5).float()    
        sort_values.append(mask.mean().item())
        fps_heatmaps[idx.item()] = heatmap.cpu()
    
    fig_images = []
    i_cluster = 0
    for i_fig in range(10):    
        fig, axs = plt.subplots(3, 5, figsize=(15, 9))
        for ax in axs.flatten():
            ax.axis("off")
        for j, idx in enumerate(fps_idx[i_fig*5:i_fig*5+5]):
            heatmap = fps_heatmaps[idx.item()]
            mask = (heatmap > 0.1).float()
            sorted_image_idxs = torch.argsort(mask.mean((1, 2)), descending=True)
            size = (images.shape[1], images.shape[2])
            heatmap = apply_reds_colormap(heatmap, size)
            for i, image_idx in enumerate(sorted_image_idxs[:3]):
                _heatmap = blend_image_with_heatmap(images[image_idx], heatmap[image_idx])
                axs[i, j].imshow(_heatmap)
                if i == 0:
                    axs[i, j].set_title(f"cluster {i_cluster+1}", fontsize=24)
                    i_cluster += 1
        plt.tight_layout(h_pad=0.5, w_pad=0.3)

        buf = BytesIO()
        plt.savefig(buf, bbox_inches='tight', dpi=72)
        
        buf.seek(0)  # Move to the start of the BytesIO buffer
        img = Image.open(buf)
        img = img.convert("RGB")
        img = copy.deepcopy(img)
        buf.close()

        fig_images.append(img)
        plt.close()
        
        # plt.imshow(img)
        # plt.axis("off")
        # plt.show()
    return fig_images

    
def ncut_run(
    model,
    images,
    model_name="DiNO(dino_vitb8_448)",
    layer=10,
    num_eig=100,
    node_type="block",
    affinity_focal_gamma=0.5,
    num_sample_ncut=10000,
    knn_ncut=10,
    embedding_method="tsne_3d",
    embedding_metric='euclidean',
    num_sample_tsne=1000,
    knn_tsne=10,
    perplexity=500,
    n_neighbors=500,
    min_dist=0.1,
    sampling_method="fps",
    old_school_ncut=False,
    recursion=False,
    recursion_l2_n_eigs=50,
    recursion_l3_n_eigs=20,
    recursion_metric="euclidean",
    recursion_l1_gamma=0.5,
    recursion_l2_gamma=0.5,
    recursion_l3_gamma=0.5,
    video_output=False,
    is_lisa=False,
    lisa_prompt1="",
    lisa_prompt2="",
    lisa_prompt3="",
):
    logging_str = ""
    if "AlignedThreeModelAttnNodes" == model_name:
        # dirty patch for the alignedcut paper
        resolution = (224, 224)
    else:
        resolution = RES_DICT[model_name]
    logging_str += f"Resolution: {resolution}\n"
    if perplexity >= num_sample_tsne or n_neighbors >= num_sample_tsne:
        # raise gr.Error("Perplexity must be less than the number of samples for t-SNE.")
        gr.Warning("Perplexity/n_neighbors must be less than the number of samples.\n" f"Setting Perplexity to {num_sample_tsne-1}.")
        logging_str += f"Perplexity/n_neighbors must be less than the number of samples.\n" f"Setting Perplexity to {num_sample_tsne-1}.\n"
        perplexity = num_sample_tsne - 1
        n_neighbors = num_sample_tsne - 1

    if torch.cuda.is_available():
        torch.cuda.empty_cache()
        
    node_type = node_type.split(":")[0].strip()
        
    start = time.time()
    if "AlignedThreeModelAttnNodes" == model_name:
        # dirty patch for the alignedcut paper
        features = run_alignedthreemodelattnnodes(images, model, batch_size=BATCH_SIZE)
    elif is_lisa == True:
        # dirty patch for the LISA model
        features = []
        with torch.no_grad():
            model = model.cuda()
            images = images.cuda()
            lisa_prompts = [lisa_prompt1, lisa_prompt2, lisa_prompt3]
            for prompt in lisa_prompts:
                import bleach
                prompt = bleach.clean(prompt)
                prompt = prompt.strip()
                # print(prompt)
                # # copy the sting to a new string
                # copy_s = copy.copy(prompt)
                feature = model(images, input_str=prompt)[node_type][0]
                feature = F.normalize(feature, dim=-1)
                features.append(feature.cpu().float())
            features = torch.stack(features)
    else:
        features = extract_features(
            images, model, node_type=node_type, layer=layer-1, batch_size=BATCH_SIZE
        )
    # print(f"Feature extraction time (gpu): {time.time() - start:.2f}s")
    logging_str += f"Backbone time: {time.time() - start:.2f}s\n"
    
    if recursion:
        rgbs = []
        recursion_gammas = [recursion_l1_gamma, recursion_l2_gamma, recursion_l3_gamma]
        inp = features
        for i, n_eigs in enumerate([num_eig, recursion_l2_n_eigs, recursion_l3_n_eigs]):
            logging_str += f"Recursion #{i+1}\n"
            rgb, _logging_str, eigvecs = compute_ncut(
                inp,
                num_eig=n_eigs,
                num_sample_ncut=num_sample_ncut,
                affinity_focal_gamma=recursion_gammas[i],
                knn_ncut=knn_ncut,
                knn_tsne=knn_tsne,
                num_sample_tsne=num_sample_tsne,
                embedding_method=embedding_method,
                embedding_metric=embedding_metric,
                perplexity=perplexity,
                n_neighbors=n_neighbors,
                min_dist=min_dist,
                sampling_method=sampling_method,
                metric="cosine" if i == 0 else recursion_metric,
            )
            logging_str += _logging_str
            rgb = dont_use_too_much_green(rgb)
            rgbs.append(to_pil_images(rgb))
            inp = eigvecs.reshape(*features.shape[:3], -1)
            if recursion_metric == "cosine":
                inp = F.normalize(inp, dim=-1)
        return rgbs[0], rgbs[1], rgbs[2], logging_str
    
    if old_school_ncut:  # individual images
        logging_str += "Running NCut for each image independently\n"
        rgb = []
        for i_image in range(features.shape[0]):
            logging_str += f"Image #{i_image+1}\n"
            feature = features[i_image]
            _rgb, _logging_str, _ = compute_ncut(
                feature[None],
                num_eig=num_eig,
                num_sample_ncut=30000,
                affinity_focal_gamma=affinity_focal_gamma,
                knn_ncut=1,
                knn_tsne=10,
                num_sample_tsne=300,
                embedding_method=embedding_method,
                embedding_metric=embedding_metric,
                perplexity=perplexity,
                n_neighbors=n_neighbors,
                min_dist=min_dist,
                sampling_method=sampling_method,
            )
            logging_str += _logging_str
            rgb.append(_rgb[0])
    
    
    cluster_images = None
    if not old_school_ncut:  # ailgnedcut, joint across all images
        rgb, _logging_str, eigvecs = compute_ncut(
            features,
            num_eig=num_eig,
            num_sample_ncut=num_sample_ncut,
            affinity_focal_gamma=affinity_focal_gamma,
            knn_ncut=knn_ncut,
            knn_tsne=knn_tsne,
            num_sample_tsne=num_sample_tsne,
            embedding_method=embedding_method,
            embedding_metric=embedding_metric,
            perplexity=perplexity,
            n_neighbors=n_neighbors,
            min_dist=min_dist,
            sampling_method=sampling_method,
        )
        logging_str += _logging_str
        
        if "AlignedThreeModelAttnNodes" == model_name:
            # dirty patch for the alignedcut paper
            # galleries = []
            # for i_node in range(rgb.shape[1]):
            #     _rgb = rgb[:, i_node]
            #     galleries.append(to_pil_images(_rgb, target_size=56))
            # return *galleries, logging_str
            pil_images = []
            for i_image in range(rgb.shape[0]):
                _im = plot_one_image_36_grid(images[i_image], rgb[i_image])
                pil_images.append(_im)
            return pil_images, logging_str
        
        
        if is_lisa == True:
            # dirty patch for the LISA model
            galleries = []
            for i_prompt in range(len(lisa_prompts)):
                _rgb = rgb[i_prompt]
                galleries.append(to_pil_images(_rgb))
            return *galleries, logging_str
        
        rgb = dont_use_too_much_green(rgb)
        
        if not video_output:
            start = time.time()
            h, w = features.shape[1], features.shape[2]
            if torch.cuda.is_available():
                images = images.cuda()
            _images = reverse_transform_image(images, stablediffusion="stable" in model_name.lower())
            cluster_images = make_cluster_plot(eigvecs, _images, h=h, w=w)
            logging_str += f"Plot time: {time.time() - start:.2f}s\n"
        
    
    if video_output:
        video_path = get_random_path()
        video_cache.add_video(video_path)
        pil_images_to_video(to_pil_images(rgb), video_path)
        return video_path, logging_str
    
    
    return to_pil_images(rgb), cluster_images, logging_str


def _ncut_run(*args, **kwargs):
    n_ret = kwargs.pop("n_ret", 1)
    try:
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
            
        ret = ncut_run(*args, **kwargs)
        
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
        
        ret = list(ret)[:n_ret] + [ret[-1]]
        return ret
    except Exception as e:
        gr.Error(str(e))
        if torch.cuda.is_available():
            torch.cuda.empty_cache()
        return *(None for _ in range(n_ret)), "Error: " + str(e)

    # ret = ncut_run(*args, **kwargs)
    # ret = list(ret)[:n_ret] + [ret[-1]]
    # return ret

if USE_HUGGINGFACE_ZEROGPU:
    @spaces.GPU(duration=20)
    def quick_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    @spaces.GPU(duration=30)
    def long_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    @spaces.GPU(duration=60)
    def longer_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    @spaces.GPU(duration=120)
    def super_duper_long_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)
    
    def cpu_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

if not USE_HUGGINGFACE_ZEROGPU:
    def quick_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    def long_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    def longer_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

    def super_duper_long_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)
    
    def cpu_run(*args, **kwargs):
        return _ncut_run(*args, **kwargs)

def extract_video_frames(video_path, max_frames=100):
    from decord import VideoReader
    vr = VideoReader(video_path)
    num_frames = len(vr)
    if num_frames > max_frames:
        gr.Warning(f"Video has {num_frames} frames. Only using {max_frames} frames. Evenly spaced.")
        frame_idx = np.linspace(0, num_frames - 1, max_frames, dtype=int).tolist()
    else:
        frame_idx = list(range(num_frames))
    frames = vr.get_batch(frame_idx).asnumpy()
    # return as list of PIL images
    return [(Image.fromarray(frames[i]), "") for i in range(frames.shape[0])]

def transform_image(image, resolution=(1024, 1024), stablediffusion=False):
    image = image.convert('RGB').resize(resolution, Image.LANCZOS)
    # Convert to torch tensor
    image = torch.tensor(np.array(image).transpose(2, 0, 1)).float()
    image = image / 255
    # Normalize
    if not stablediffusion:
        mean = [0.485, 0.456, 0.406]
        std = [0.229, 0.224, 0.225]
        image = (image - torch.tensor(mean).view(3, 1, 1)) / torch.tensor(std).view(3, 1, 1)
    if stablediffusion:
        image = image * 2 - 1
    return image

def reverse_transform_image(image, stablediffusion=False):
    if stablediffusion:
        image = (image + 1) / 2
    else:
        mean = torch.tensor([0.485, 0.456, 0.406]).view(3, 1, 1).to(image.device)
        std = torch.tensor([0.229, 0.224, 0.225]).view(3, 1, 1).to(image.device)
        image = image * std + mean
    image = torch.clamp(image, 0, 1)
    return image

def plot_one_image_36_grid(original_image, tsne_rgb_images):
    mean = [0.485, 0.456, 0.406]
    std = [0.229, 0.224, 0.225]
    original_image = original_image * torch.tensor(std).view(3, 1, 1) + torch.tensor(mean).view(3, 1, 1)
    original_image = torch.clamp(original_image, 0, 1)
    
    fig = plt.figure(figsize=(20, 4))
    grid = plt.GridSpec(3, 14, hspace=0.1, wspace=0.1)

    ax1 = fig.add_subplot(grid[0:2, 0:2])
    img = original_image.cpu().float().numpy().transpose(1, 2, 0)

    def convert_and_pad_image(np_array, pad_size=20):
        """
        Converts a NumPy array of shape (height, width, 3) to a PNG image
        and pads the right and bottom sides with a transparent background.

        Args:
            np_array (numpy.ndarray): Input NumPy array of shape (height, width, 3)
            pad_size (int, optional): Number of pixels to pad on the right and bottom sides. Default is 20.

        Returns:
            PIL.Image: Padded PNG image with transparent background
        """
        # Convert NumPy array to PIL Image
        img = Image.fromarray(np_array)

        # Get the original size
        width, height = img.size

        # Create a new image with padding and transparent background
        new_width = width + pad_size
        new_height = height + pad_size
        padded_img = Image.new('RGBA', (new_width, new_height), color=(255, 255, 255, 0))

        # Paste the original image onto the padded image
        padded_img.paste(img, (0, 0))

        return padded_img
    
    img = convert_and_pad_image((img*255).astype(np.uint8))
    ax1.imshow(img)
    ax1.axis('off')

    model_names = ['CLIP', 'DINO', 'MAE']

    for i_model, model_name in enumerate(model_names):
        for i_layer in range(12):
            ax = fig.add_subplot(grid[i_model, i_layer+2])
            ax.imshow(tsne_rgb_images[i_layer+12*i_model].cpu().float().numpy())
            ax.axis('off')
            if i_model == 0:
                ax.set_title(f'Layer{i_layer}', fontsize=16)
            if i_layer == 0:
                ax.text(-0.1, 0.5, model_name, va="center", ha="center", fontsize=16, transform=ax.transAxes, rotation=90,)    
    plt.tight_layout()
    buf = BytesIO()
    plt.savefig(buf, bbox_inches='tight', pad_inches=0, dpi=100)
    
    buf.seek(0)  # Move to the start of the BytesIO buffer
    img = Image.open(buf)
    img = img.convert("RGB")
    img = copy.deepcopy(img)
    buf.close()
    plt.close()
    return img

def load_alignedthreemodel():
    
    os.system("git clone https://huggingface.co/huzey/alignedthreeattn >> /dev/null 2>&1")
    # pull
    os.system("git -C alignedthreeattn pull >> /dev/null 2>&1")
    # add to path
    import sys
    sys.path.append("alignedthreeattn")
    
    
    from alignedthreeattn.alignedthreeattn_model import ThreeAttnNodes
    
    align_weights = torch.load("alignedthreeattn/align_weights.pth")
    model = ThreeAttnNodes(align_weights)
    
    # url = 'https://huggingface.co/huzey/aligned_model_test/resolve/main/3attn_nodes.pth'
    # save_path = "alignedthreemodel.pth"
    # if not os.path.exists(save_path):
    #     os.system(f"wget {url} -O {save_path} -q")
    # model = torch.load(save_path)
    return model

promptable_diffusion_models = ["Diffusion(stabilityai/stable-diffusion-2)", "Diffusion(CompVis/stable-diffusion-v1-4)"]
promptable_segmentation_models = ["LISA(xinlai/LISA-7B-v1)"]


def run_fn(
    images,
    model_name="DiNO(dino_vitb8_448)",
    layer=10,
    num_eig=100,
    node_type="block",
    positive_prompt="",
    negative_prompt="",
    is_lisa=False,
    lisa_prompt1="",
    lisa_prompt2="",
    lisa_prompt3="",
    affinity_focal_gamma=0.5,
    num_sample_ncut=10000,
    knn_ncut=10,
    embedding_method="tsne_3d",
    embedding_metric='euclidean',
    num_sample_tsne=300,
    knn_tsne=10,
    perplexity=150,
    n_neighbors=150,
    min_dist=0.1,
    sampling_method="fps",
    old_school_ncut=False,
    max_frames=100,
    recursion=False,
    recursion_l2_n_eigs=50,
    recursion_l3_n_eigs=20,
    recursion_metric="euclidean",
    recursion_l1_gamma=0.5,
    recursion_l2_gamma=0.5,
    recursion_l3_gamma=0.5,
    n_ret=1,
):
    
    if images is None:
        gr.Warning("No images selected.")
        return *(None for _ in range(n_ret)), "No images selected."
    
    video_output = False
    if isinstance(images, str):
        images = extract_video_frames(images, max_frames=max_frames)
        video_output = True
    
    if sampling_method == "fps":
        sampling_method = "farthest"
        
    # resize the images before acquiring GPU
    if "AlignedThreeModelAttnNodes" == model_name:
        # dirty patch for the alignedcut paper
        resolution = (224, 224)
    else:
        resolution = RES_DICT[model_name]
    images = [tup[0] for tup in images]
    stablediffusion = True if "Diffusion" in model_name else False
    images = [transform_image(image, resolution=resolution, stablediffusion=stablediffusion) for image in images]
    images = torch.stack(images)
    
    
    if is_lisa:
        import subprocess
        import sys
        import importlib
        gr.Warning("LISA model is not compatible with the current version of transformers. Please contact the LISA and Llava author for update.")
        gr.Warning("This is a dirty patch for the LISA model. switch to the old version of transformers.")
        gr.Warning("Not garanteed to work.")
        # LISA and Llava is not compatible with the current version of transformers
        # please contact the author for update
        # this is a dirty patch for the LISA model
        
        # pre-import the SD3 pipeline
        from diffusers import StableDiffusion3Pipeline
        
        # unloading the current transformers
        for module in list(sys.modules.keys()):
            if "transformers" in module:
                del sys.modules[module]
            

        def install_transformers_version(version, target_dir):
            """Install a specific version of transformers to a target directory."""
            if not os.path.exists(target_dir):
                os.makedirs(target_dir)
            
            # Use subprocess to run the pip command
            # subprocess.check_call([sys.executable, '-m', 'pip', 'install', f'transformers=={version}', '-t', target_dir])
            os.system(f"{sys.executable} -m pip install transformers=={version} -t {target_dir} >> /dev/null 2>&1")

        target_dir = '/tmp/lisa_transformers_v433'
        if not os.path.exists(target_dir):
            install_transformers_version('4.33.0', target_dir)

        # Add the new version path to sys.path
        sys.path.insert(0, target_dir)
        
        transformers = importlib.import_module("transformers")

    if not is_lisa:
        import subprocess
        import sys
        import importlib
        # remove the LISA model from the sys.path
            
        if "/tmp/lisa_transformers_v433" in sys.path:
            sys.path.remove("/tmp/lisa_transformers_v433")
        
        transformers = importlib.import_module("transformers")
        
    
    
    if "AlignedThreeModelAttnNodes" == model_name:
        # dirty patch for the alignedcut paper
        model = load_alignedthreemodel()
    else:
        model = load_model(model_name)
        
    if "stable" in model_name.lower() and "diffusion" in model_name.lower():
        model.timestep = layer
        layer = 1
        
    if model_name in promptable_diffusion_models:
        model.positive_prompt = positive_prompt
        model.negative_prompt = negative_prompt
        
    kwargs = {
        "model_name": model_name,
        "layer": layer,
        "num_eig": num_eig,
        "node_type": node_type,
        "affinity_focal_gamma": affinity_focal_gamma,
        "num_sample_ncut": num_sample_ncut,
        "knn_ncut": knn_ncut,
        "embedding_method": embedding_method,
        "embedding_metric": embedding_metric,
        "num_sample_tsne": num_sample_tsne,
        "knn_tsne": knn_tsne,
        "perplexity": perplexity,
        "n_neighbors": n_neighbors,
        "min_dist": min_dist,
        "sampling_method": sampling_method,
        "old_school_ncut": old_school_ncut,
        "recursion": recursion,
        "recursion_l2_n_eigs": recursion_l2_n_eigs,
        "recursion_l3_n_eigs": recursion_l3_n_eigs,
        "recursion_metric": recursion_metric,
        "recursion_l1_gamma": recursion_l1_gamma,
        "recursion_l2_gamma": recursion_l2_gamma,
        "recursion_l3_gamma": recursion_l3_gamma,
        "video_output": video_output,
        "lisa_prompt1": lisa_prompt1,
        "lisa_prompt2": lisa_prompt2,
        "lisa_prompt3": lisa_prompt3,
        "is_lisa": is_lisa,
        "n_ret": n_ret,
    }
    # print(kwargs)
    
    if old_school_ncut:
        return super_duper_long_run(model, images, **kwargs)
    
    if is_lisa:
        return super_duper_long_run(model, images, **kwargs)
    
    num_images = len(images)
    if num_images >= 100:
        return super_duper_long_run(model, images, **kwargs)
    if 'diffusion' in model_name.lower():
        return super_duper_long_run(model, images, **kwargs)
    if recursion:
        return longer_run(model, images, **kwargs)
    if num_images >= 50:
        return longer_run(model, images, **kwargs)
    if old_school_ncut:
        return longer_run(model, images, **kwargs)
    if num_images >= 10:
        return long_run(model, images, **kwargs)
    if embedding_method == "UMAP":
        if perplexity >= 250 or num_sample_tsne >= 500:
            return longer_run(model, images, **kwargs)
        return long_run(model, images, **kwargs)
    if embedding_method == "t-SNE":
        if perplexity >= 250 or num_sample_tsne >= 500:
            return long_run(model, images, **kwargs)
        return quick_run(model, images, **kwargs)
    
    return quick_run(model, images, **kwargs)



def make_input_images_section():
    gr.Markdown('### Input Images')
    input_gallery = gr.Gallery(value=None, label="Select images", show_label=False, elem_id="images", columns=[3], rows=[1], object_fit="contain", height="auto", type="pil", show_share_button=False)
    submit_button = gr.Button("🔴 RUN", elem_id="submit_button", variant='primary')
    clear_images_button = gr.Button("🗑️Clear", elem_id='clear_button', variant='stop')
    return input_gallery, submit_button, clear_images_button

def make_input_video_section():
    # gr.Markdown('### Input Video')
    input_gallery = gr.Video(value=None, label="Select video", elem_id="video-input", height="auto", show_share_button=False)
    gr.Markdown('_image backbone model is used to extract features from each frame, NCUT is computed on all frames_')
    # max_frames_number = gr.Number(100, label="Max frames", elem_id="max_frames")
    max_frames_number = gr.Slider(1, 200, step=1, label="Max frames", value=100, elem_id="max_frames")
    submit_button = gr.Button("🔴 RUN", elem_id="submit_button", variant='primary')
    clear_images_button = gr.Button("🗑️Clear", elem_id='clear_button', variant='stop')
    return input_gallery, submit_button, clear_images_button, max_frames_number

def make_dataset_images_section(advanced=False, is_random=False):

    gr.Markdown('### Load Datasets')
    load_images_button = gr.Button("🔴 Load Images", elem_id="load-images-button", variant='primary')
    advanced_radio = gr.Radio(["Basic", "Advanced"], label="Datasets", value="Advanced" if advanced else "Basic", elem_id="advanced-radio")
    with gr.Column() as basic_block:
        example_gallery = gr.Gallery(value=example_items, label="Example Set A", show_label=False, columns=[3], rows=[2], object_fit="scale-down", height="200px", show_share_button=False, elem_id="example-gallery")
    with gr.Column() as advanced_block:
        dataset_names = DATASET_NAMES
        dataset_classes = DATASET_CLASSES
        with gr.Row():
            dataset_dropdown = gr.Dropdown(dataset_names, label="Dataset name", value="mrm8488/ImageNet1K-val", elem_id="dataset", min_width=300)
            num_images_slider = gr.Number(10, label="Number of images", elem_id="num_images")
            if not is_random:
                filter_by_class_checkbox = gr.Checkbox(label="Filter by class", value=True, elem_id="filter_by_class_checkbox")
                filter_by_class_text = gr.Textbox(label="Class to select", value="0,33,99", elem_id="filter_by_class_text", info=f"e.g. `0,1,2`. (1000 classes)", visible=True)
                is_random_checkbox = gr.Checkbox(label="Random shuffle", value=False, elem_id="random_seed_checkbox")
                random_seed_slider = gr.Slider(0, 1000, step=1, label="Random seed", value=1, elem_id="random_seed", visible=False)
            if is_random:
                filter_by_class_checkbox = gr.Checkbox(label="Filter by class", value=False, elem_id="filter_by_class_checkbox")
                filter_by_class_text = gr.Textbox(label="Class to select", value="0,33,99", elem_id="filter_by_class_text", info=f"e.g. `0,1,2`. (1000 classes)", visible=False)
                is_random_checkbox = gr.Checkbox(label="Random shuffle", value=True, elem_id="random_seed_checkbox")
                random_seed_slider = gr.Slider(0, 1000, step=1, label="Random seed", value=42, elem_id="random_seed", visible=True)
                
    
    if advanced:
        advanced_block.visible = True
        basic_block.visible = False
    else:
        advanced_block.visible = False
        basic_block.visible = True
        
    # change visibility 
    advanced_radio.change(fn=lambda x: gr.update(visible=x=="Advanced"), inputs=advanced_radio, outputs=[advanced_block])
    advanced_radio.change(fn=lambda x: gr.update(visible=x=="Basic"), inputs=advanced_radio, outputs=[basic_block])
    
    
    def change_filter_options(dataset_name):
        idx = dataset_names.index(dataset_name)
        num_classes = dataset_classes[idx]
        if num_classes is None:
            return (gr.Checkbox(label="Filter by class", value=False, elem_id="filter_by_class_checkbox", visible=False),
                    gr.Textbox(label="Class to select", value="0,1,2", elem_id="filter_by_class_text", info="e.g. `0,1,2`. This dataset has no class label", visible=False))
        return (gr.Checkbox(label="Filter by class", value=True, elem_id="filter_by_class_checkbox", visible=True),
                gr.Textbox(label="Class to select", value="0,1,2", elem_id="filter_by_class_text", info=f"e.g. `0,1,2`. ({num_classes} classes)", visible=True))
    dataset_dropdown.change(fn=change_filter_options, inputs=dataset_dropdown, outputs=[filter_by_class_checkbox, filter_by_class_text])
    
    def change_filter_by_class(is_filter, dataset_name):
        idx = dataset_names.index(dataset_name)
        num_classes = dataset_classes[idx]
        return gr.Textbox(label="Class to select", value="0,1,2", elem_id="filter_by_class_text", info=f"e.g. `0,1,2`. ({num_classes} classes)", visible=is_filter)
    filter_by_class_checkbox.change(fn=change_filter_by_class, inputs=[filter_by_class_checkbox, dataset_dropdown], outputs=filter_by_class_text)
    
    def change_random_seed(is_random):
        return gr.Slider(0, 1000, step=1, label="Random seed", value=1, elem_id="random_seed", visible=is_random)
    is_random_checkbox.change(fn=change_random_seed, inputs=is_random_checkbox, outputs=random_seed_slider)
    
        
    def load_dataset_images(is_advanced, dataset_name, num_images=10, 
                            is_filter=True, filter_by_class_text="0,1,2", 
                            is_random=False, seed=1):
        if is_advanced == "Basic":
            gr.Info("Loaded images from Ego-Exo4D")
            return default_images
        try:
            dataset = load_dataset(dataset_name, trust_remote_code=True)
            key = list(dataset.keys())[0]
            dataset = dataset[key]
        except Exception as e:
            gr.Error(f"Error loading dataset {dataset_name}: {e}")
            return None
        if num_images > len(dataset):
            num_images = len(dataset)
        
        if is_filter:
            classes = [int(i) for i in filter_by_class_text.split(",")]
            labels = np.array(dataset['label'])
            unique_labels = np.unique(labels)
            valid_classes = [i for i in classes if i in unique_labels]
            invalid_classes = [i for i in classes if i not in unique_labels]
            if len(invalid_classes) > 0:
                gr.Warning(f"Classes {invalid_classes} not found in the dataset.")
            if len(valid_classes) == 0:
                gr.Error(f"Classes {classes} not found in the dataset.")
                return None
            # shuffle each class
            chunk_size = num_images // len(valid_classes)
            image_idx = []
            for i in valid_classes:
                idx = np.where(labels == i)[0]
                if is_random:
                    idx = np.random.RandomState(seed).choice(idx, chunk_size, replace=False)
                else:
                    idx = idx[:chunk_size]
                image_idx.extend(idx.tolist())
        if not is_filter:
            if is_random:
                image_idx = np.random.RandomState(seed).choice(len(dataset), num_images, replace=False).tolist()
            else:
                image_idx = list(range(num_images))
        images = [dataset[i]['image'] for i in image_idx]
        gr.Info(f"Loaded {len(images)} images from {dataset_name}")
        return images   
    
    load_images_button.click(load_dataset_images, 
                        inputs=[advanced_radio, dataset_dropdown, num_images_slider,
                                filter_by_class_checkbox, filter_by_class_text, 
                                is_random_checkbox, random_seed_slider],
                        outputs=[input_gallery])
    
    return dataset_dropdown, num_images_slider, random_seed_slider, load_images_button
    

def random_rotate_rgb_gallery(images):
    if images is None or len(images) == 0:
        gr.Warning("No images selected.")
        return []
    # read webp images
    images = [Image.open(image[0]).convert("RGB") for image in images]
    images = [np.array(image).astype(np.float32) for image in images]
    images = np.stack(images)
    images = torch.tensor(images) / 255
    position = np.random.choice([1, 2, 4, 5, 6])
    images = rotate_rgb_cube(images, position)
    images = to_pil_images(images, resize=False)
    return images

def sequence_rotate_rgb_gallery(images):
    if images is None or len(images) == 0:
        gr.Warning("No images selected.")
        return []
    # read webp images
    images = [Image.open(image[0]).convert("RGB") for image in images]
    images = [np.array(image).astype(np.float32) for image in images]
    images = np.stack(images)
    images = torch.tensor(images) / 255
    rotation_matrix = torch.tensor([[0, 1, 0], [0, 0, 1], [1, 0, 0]]).float()
    images = images @ rotation_matrix
    images = to_pil_images(images, resize=False)
    return images

def flip_rgb_gallery(images, axis=0):
    if images is None or len(images) == 0:
        gr.Warning("No images selected.")
        return []
    # read webp images
    images = [Image.open(image[0]).convert("RGB") for image in images]
    images = [np.array(image).astype(np.float32) for image in images]
    images = np.stack(images)
    images = torch.tensor(images) / 255
    images = 1 - images
    images = to_pil_images(images, resize=False)
    return images

def add_output_images_buttons(output_gallery):
    with gr.Row():
        rotate_button = gr.Button("🔄 Rotate", elem_id="rotate_button", variant='secondary')
        rotate_button.click(sequence_rotate_rgb_gallery, inputs=[output_gallery], outputs=[output_gallery])
        flip_button = gr.Button("🔃 Flip", elem_id="flip_button", variant='secondary')
        flip_button.click(flip_rgb_gallery, inputs=[output_gallery], outputs=[output_gallery])
    return rotate_button, flip_button
    

def make_output_images_section():
    gr.Markdown('### Output Images')
    output_gallery = gr.Gallery(format='png', value=[], label="NCUT Embedding", show_label=False, elem_id="ncut", columns=[3], rows=[1], object_fit="contain", height="auto", show_share_button=True)
    add_output_images_buttons(output_gallery)
    return output_gallery

def make_parameters_section(is_lisa=False):
    gr.Markdown("### Parameters <a style='color: #0044CC;' href='https://ncut-pytorch.readthedocs.io/en/latest/how_to_get_better_segmentation/' target='_blank'>Help</a>")
    from ncut_pytorch.backbone import list_models, get_demo_model_names
    model_names = list_models()
    model_names = sorted(model_names)
    if is_lisa:
        model_dropdown = gr.Dropdown(["LISA(xinlai/LISA-7B-v1)"], label="Backbone", value="LISA(xinlai/LISA-7B-v1)", elem_id="model_name")
        layer_slider = gr.Slider(1, 6, step=1, label="LISA decoder: Layer index", value=6, elem_id="layer", visible=False)        
        layer_names = ["dec_0_input", "dec_0_attn", "dec_0_block", "dec_1_input", "dec_1_attn", "dec_1_block"]
        positive_prompt = gr.Textbox(label="Prompt (Positive)", elem_id="prompt", placeholder="e.g. 'a photo of Gibson Les Pual guitar'", visible=False)
        negative_prompt = gr.Textbox(label="Prompt (Negative)", elem_id="prompt", placeholder="e.g. 'a photo from egocentric view'", visible=False)
        node_type_dropdown = gr.Dropdown(layer_names, label="LISA (SAM) decoder: Layer and Node", value="dec_1_block", elem_id="node_type")
    else:
        # remove LISA from the list
        model_names = [m for m in model_names if "LISA" not in m]
        model_dropdown = gr.Dropdown(model_names, label="Backbone", value="DiNO(dino_vitb8_448)", elem_id="model_name")
        layer_slider = gr.Slider(1, 12, step=1, label="Backbone: Layer index", value=10, elem_id="layer")
        positive_prompt = gr.Textbox(label="Prompt (Positive)", elem_id="prompt", placeholder="e.g. 'a photo of Gibson Les Pual guitar'")
        positive_prompt.visible = False
        negative_prompt = gr.Textbox(label="Prompt (Negative)", elem_id="prompt", placeholder="e.g. 'a photo from egocentric view'")
        negative_prompt.visible = False
        node_type_dropdown = gr.Dropdown(["attn: attention output", "mlp: mlp output", "block: sum of residual"], label="Backbone: Layer type", value="block: sum of residual", elem_id="node_type", info="which feature to take from each layer?")
    num_eig_slider = gr.Slider(1, 1000, step=1, label="NCUT: Number of eigenvectors", value=100, elem_id="num_eig", info='increase for more clusters')

    def change_layer_slider(model_name):
        # SD2, UNET
        if "stable" in model_name.lower() and "diffusion" in model_name.lower():
            from ncut_pytorch.backbone import SD_KEY_DICT
            default_layer = 'up_2_resnets_1_block' if 'diffusion-3' not in model_name else 'block_23'
            return (gr.Slider(1, 49, step=1, label="Diffusion: Timestep (Noise)", value=5, elem_id="layer", visible=True, info="Noise level, 50 is max noise"),
                    gr.Dropdown(SD_KEY_DICT[model_name], label="Diffusion: Layer and Node", value=default_layer, elem_id="node_type", info="U-Net (v1, v2) or DiT (v3)"))
        
        if model_name == "LISSL(xinlai/LISSL-7B-v1)":
            layer_names = ["dec_0_input", "dec_0_attn", "dec_0_block", "dec_1_input", "dec_1_attn", "dec_1_block"]
            default_layer = "dec_1_block"
            return (gr.Slider(1, 6, step=1, label="LISA decoder: Layer index", value=6, elem_id="layer", visible=False, info=""),
                    gr.Dropdown(layer_names, label="LISA decoder: Layer and Node", value=default_layer, elem_id="node_type"))

        layer_dict = LAYER_DICT
        if model_name in layer_dict:
            value = layer_dict[model_name]
            return (gr.Slider(1, value, step=1, label="Backbone: Layer index", value=value, elem_id="layer", visible=True, info=""),
                    gr.Dropdown(["attn: attention output", "mlp: mlp output", "block: sum of residual"], label="Backbone: Layer type", value="block: sum of residual", elem_id="node_type", info="which feature to take from each layer?"))
        else:
            value = 12
            return (gr.Slider(1, value, step=1, label="Backbone: Layer index", value=value, elem_id="layer", visible=True, info=""),
                    gr.Dropdown(["attn: attention output", "mlp: mlp output", "block: sum of residual"], label="Backbone: Layer type", value="block: sum of residual", elem_id="node_type", info="which feature to take from each layer?"))
    model_dropdown.change(fn=change_layer_slider, inputs=model_dropdown, outputs=[layer_slider, node_type_dropdown])
    
    def change_prompt_text(model_name):
        if model_name in promptable_diffusion_models:
            return (gr.Textbox(label="Prompt (Positive)", elem_id="prompt", placeholder="e.g. 'a photo of Gibson Les Pual guitar'", visible=True),
                    gr.Textbox(label="Prompt (Negative)", elem_id="prompt", placeholder="e.g. 'a photo from egocentric view'", visible=True))
        return (gr.Textbox(label="Prompt (Positive)", elem_id="prompt", placeholder="e.g. 'a photo of Gibson Les Pual guitar'", visible=False),
                gr.Textbox(label="Prompt (Negative)", elem_id="prompt", placeholder="e.g. 'a photo from egocentric view'", visible=False))
    model_dropdown.change(fn=change_prompt_text, inputs=model_dropdown, outputs=[positive_prompt, negative_prompt])
    
    with gr.Accordion("➡️ Click to expand: more parameters", open=False):
        gr.Markdown("<a href='https://ncut-pytorch.readthedocs.io/en/latest/how_to_get_better_segmentation/' target='_blank'>Docs: How to Get Better Segmentation</a>")
        affinity_focal_gamma_slider = gr.Slider(0.01, 1, step=0.01, label="NCUT: Affinity focal gamma", value=0.5, elem_id="affinity_focal_gamma", info="decrease for shaper segmentation")
        num_sample_ncut_slider = gr.Slider(100, 50000, step=100, label="NCUT: num_sample", value=10000, elem_id="num_sample_ncut", info="Nyström approximation")
        sampling_method_dropdown = gr.Dropdown(["fps", "random"], label="NCUT: Sampling method", value="fps", elem_id="sampling_method", info="Nyström approximation")
        knn_ncut_slider = gr.Slider(1, 100, step=1, label="NCUT: KNN", value=10, elem_id="knn_ncut", info="Nyström approximation")
        embedding_method_dropdown = gr.Dropdown(["tsne_3d", "umap_3d", "umap_shpere", "tsne_2d", "umap_2d"], label="Coloring method", value="tsne_3d", elem_id="embedding_method")
        embedding_metric_dropdown = gr.Dropdown(["euclidean", "cosine"], label="t-SNE/UMAP metric", value="euclidean", elem_id="embedding_metric")
        num_sample_tsne_slider = gr.Slider(100, 10000, step=100, label="t-SNE/UMAP: num_sample", value=300, elem_id="num_sample_tsne", info="Nyström approximation")
        knn_tsne_slider = gr.Slider(1, 100, step=1, label="t-SNE/UMAP: KNN", value=10, elem_id="knn_tsne", info="Nyström approximation")
        perplexity_slider = gr.Slider(10, 1000, step=10, label="t-SNE: perplexity", value=150, elem_id="perplexity")
        n_neighbors_slider = gr.Slider(10, 1000, step=10, label="UMAP: n_neighbors", value=150, elem_id="n_neighbors")
        min_dist_slider = gr.Slider(0.1, 1, step=0.1, label="UMAP: min_dist", value=0.1, elem_id="min_dist")
    return [model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
            affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
            embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
            perplexity_slider, n_neighbors_slider, min_dist_slider, 
            sampling_method_dropdown, positive_prompt, negative_prompt]

demo = gr.Blocks(
    theme=gr.themes.Base(spacing_size='md', text_size='lg', primary_hue='blue', neutral_hue='slate', secondary_hue='pink'),
    # fill_width=False,
    # title="ncut-pytorch",
)
with demo:
    with gr.Tab('AlignedCut'):

        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()
                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section()
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
                
            with gr.Column(scale=5, min_width=200):
                output_gallery = make_output_images_section()
                cluster_gallery = gr.Gallery(value=[], label="Clusters", show_label=False, elem_id="clusters", columns=[2], rows=[1], object_fit="contain", height=450, show_share_button=True, preview=True)
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                num_eig_slider.value = 30

        clear_images_button.click(lambda x: ([], [], []), outputs=[input_gallery, output_gallery, cluster_gallery])
        
        false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
        no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
        
        submit_button.click(
            partial(run_fn, n_ret=2),
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                false_placeholder, no_prompt, no_prompt, no_prompt,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown
            ],
            outputs=[output_gallery, cluster_gallery, logging_text],
            api_name="API_AlignedCut"
        )
        
    with gr.Tab('NCut'): 
        gr.Markdown('#### NCut (Legacy), not aligned, no Nyström approximation')
        gr.Markdown('Each image is solved independently, <em>color is <b>not</b> aligned across images</em>')
        
        gr.Markdown('---')
        gr.Markdown('<p style="text-align: center;"><b>NCut    vs.   AlignedCut</b></p>')
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('#### Pros')
                gr.Markdown('- Easy Solution. Use less eigenvectors.')
                gr.Markdown('- Exact solution. No Nyström approximation.')
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('#### Cons')
                gr.Markdown('- Not aligned. Distance is not preserved across images. No pseudo-labeling or correspondence.')
                gr.Markdown('- Poor complexity scaling. Unable to handle large number of pixels.')
        gr.Markdown('---')
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                gr.Markdown(' ')
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('<em>color is <b>not</b> aligned across images</em> 👇')


        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()
                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section()
                
            with gr.Column(scale=5, min_width=200):
                output_gallery = make_output_images_section()
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                old_school_ncut_checkbox = gr.Checkbox(label="Old school NCut", value=True, elem_id="old_school_ncut")
                invisible_list = [old_school_ncut_checkbox, num_sample_ncut_slider, knn_ncut_slider,
                                    num_sample_tsne_slider, knn_tsne_slider, sampling_method_dropdown]
                for item in invisible_list:
                    item.visible = False
                # logging text box
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
            
        clear_images_button.click(lambda x: ([], []), outputs=[input_gallery, output_gallery])
        false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
        no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
        
        submit_button.click(
            run_fn,
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                false_placeholder, no_prompt, no_prompt, no_prompt,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown, 
                old_school_ncut_checkbox
            ],
            outputs=[output_gallery, logging_text],
            api_name="API_NCut",
        )
        
    with gr.Tab('Recursive Cut'): 
        gr.Markdown('NCUT can be applied recursively, the eigenvectors from previous iteration is the input for the next iteration NCUT. ')
        gr.Markdown('__Recursive NCUT__ amplifies small object parts, please see [Documentation](https://ncut-pytorch.readthedocs.io/en/latest/how_to_get_better_segmentation/#recursive-ncut)')
                
        gr.Markdown('---')

        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Recursion #1)')
                l1_gallery = gr.Gallery(format='png', value=[], label="Recursion #1", show_label=False, elem_id="ncut_l1", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                add_output_images_buttons(l1_gallery)
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Recursion #2)')
                l2_gallery = gr.Gallery(format='png', value=[], label="Recursion #2", show_label=False, elem_id="ncut_l2", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                add_output_images_buttons(l2_gallery)
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Recursion #3)')
                l3_gallery = gr.Gallery(format='png', value=[], label="Recursion #3", show_label=False, elem_id="ncut_l3", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                add_output_images_buttons(l3_gallery)
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()
                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section(advanced=True)
                num_images_slider.value = 100
                clear_images_button.visible = False
                
            with gr.Column(scale=5, min_width=200):
                with gr.Accordion("➡️ Recursion config", open=True):
                    l1_num_eig_slider = gr.Slider(1, 1000, step=1, label="Recursion #1: N eigenvectors", value=100, elem_id="l1_num_eig")
                    l2_num_eig_slider = gr.Slider(1, 1000, step=1, label="Recursion #2: N eigenvectors", value=50, elem_id="l2_num_eig")
                    l3_num_eig_slider = gr.Slider(1, 1000, step=1, label="Recursion #3: N eigenvectors", value=50, elem_id="l3_num_eig")
                    metric_dropdown = gr.Dropdown(["euclidean", "cosine"], label="Recursion distance metric", value="cosine", elem_id="recursion_metric")
                    l1_affinity_focal_gamma_slider = gr.Slider(0.01, 1, step=0.01, label="Recursion #1: Affinity focal gamma", value=0.5, elem_id="recursion_l1_gamma")
                    l2_affinity_focal_gamma_slider = gr.Slider(0.01, 1, step=0.01, label="Recursion #2: Affinity focal gamma", value=0.5, elem_id="recursion_l2_gamma")
                    l3_affinity_focal_gamma_slider = gr.Slider(0.01, 1, step=0.01, label="Recursion #3: Affinity focal gamma", value=0.5, elem_id="recursion_l3_gamma")
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                num_eig_slider.visible = False
                affinity_focal_gamma_slider.visible = False
                # logging text box
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                gr.Markdown(' ')
            with gr.Column(scale=5, min_width=200):
                gr.Markdown(' ')
            with gr.Column(scale=5, min_width=200):
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
        true_placeholder = gr.Checkbox(label="True placeholder", value=True, elem_id="true_placeholder")
        true_placeholder.visible = False
        false_placeholder = gr.Checkbox(label="False placeholder", value=False, elem_id="false_placeholder")
        false_placeholder.visible = False
        number_placeholder = gr.Number(0, label="Number placeholder", elem_id="number_placeholder")
        number_placeholder.visible = False
        clear_images_button.click(lambda x: ([],), outputs=[input_gallery])
        false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
        no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
        
        submit_button.click(
            partial(run_fn, n_ret=3),
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                false_placeholder, no_prompt, no_prompt, no_prompt,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown,
                false_placeholder, number_placeholder, true_placeholder,
                l2_num_eig_slider, l3_num_eig_slider, metric_dropdown, 
                l1_affinity_focal_gamma_slider, l2_affinity_focal_gamma_slider, l3_affinity_focal_gamma_slider
            ],
            outputs=[l1_gallery, l2_gallery, l3_gallery, logging_text],
            api_name="API_RecursiveCut"
        )

        
    with gr.Tab('Video'): 
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                video_input_gallery, submit_button, clear_video_button, max_frame_number = make_input_video_section()
            with gr.Column(scale=5, min_width=200):
                video_output_gallery = gr.Video(value=None, label="NCUT Embedding", elem_id="ncut", height="auto", show_share_button=False)
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                num_sample_tsne_slider.value = 1000
                perplexity_slider.value = 500
                n_neighbors_slider.value = 500
                knn_tsne_slider.value = 20
                # logging text box
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
        clear_video_button.click(lambda x: (None, None), outputs=[video_input_gallery, video_output_gallery])
        place_holder_false = gr.Checkbox(label="Place holder", value=False, elem_id="place_holder_false")
        place_holder_false.visible = False
        false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
        no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
        
        submit_button.click(
            run_fn,
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                false_placeholder, no_prompt, no_prompt, no_prompt,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown,
                place_holder_false, max_frame_number
            ],
            outputs=[video_output_gallery, logging_text],
            api_name="API_VideoCut",
        )    
    
    with gr.Tab('Text'): 
        try:
            from app_text import make_demo
        except ImportError:
            print("Debugging")
            from draft_gradio_app_text import make_demo
        make_demo()
        
    with gr.Tab('Vision-Language'):
        gr.Markdown('[LISA](https://arxiv.org/pdf/2308.00692) is a vision-language model. Input a text prompt and image, LISA generate segmentation masks.')
        gr.Markdown('In the mask decoder layers, LISA updates the image features w.r.t. the text prompt')
        gr.Markdown('This page aims to see how the text prompt affects the image features')
        gr.Markdown('---')
        gr.Markdown('<p style="text-align: center;">Color is <b>aligned</b> across 3 prompts. NCUT is computed on the concatenated features from 3 prompts.</p>')
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Prompt #1)')
                l1_gallery = gr.Gallery(format='png', value=[], label="Prompt #1", show_label=False, elem_id="ncut_p1", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                prompt1 = gr.Textbox(label="Input Prompt #1", elem_id="prompt1", value="where is the person, include the clothes, don't include the guitar and chair", lines=3)
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Prompt #2)')
                l2_gallery = gr.Gallery(format='png', value=[], label="Prompt #2", show_label=False, elem_id="ncut_p2", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                prompt2 = gr.Textbox(label="Input Prompt #2", elem_id="prompt2", value="where is the Gibson Les Pual guitar", lines=3)
            with gr.Column(scale=5, min_width=200):
                gr.Markdown('### Output (Prompt #3)')
                l3_gallery = gr.Gallery(format='png', value=[], label="Prompt #3", show_label=False, elem_id="ncut_p3", columns=[3], rows=[5], object_fit="contain", height="auto", show_fullscreen_button=True)
                prompt3 = gr.Textbox(label="Input Prompt #3", elem_id="prompt3", value="where is the floor", lines=3)
                
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()
                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section(advanced=False)
                clear_images_button.click(lambda x: ([], [], [], []), outputs=[input_gallery, l1_gallery, l2_gallery, l3_gallery])
                
            with gr.Column(scale=5, min_width=200):
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section(is_lisa=True)
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
        
        galleries = [l1_gallery, l2_gallery, l3_gallery]
        true_placeholder = gr.Checkbox(label="True placeholder", value=True, elem_id="true_placeholder", visible=False)
        submit_button.click(
            partial(run_fn, n_ret=len(galleries)),
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                true_placeholder, prompt1, prompt2, prompt3,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown
            ],
            outputs=galleries + [logging_text],
        )
                
    with gr.Tab('Model Aligned'): 
        gr.Markdown('This page reproduce the results from the paper [AlignedCut](https://arxiv.org/abs/2406.18344)')
        gr.Markdown('---')
        gr.Markdown('**Features are aligned across models and layers.** A linear alignment transform is trained for each model/layer, learning signal comes from 1) fMRI brain activation and 2) segmentation preserving eigen-constraints.')
        gr.Markdown('NCUT is computed on the concatenated graph of all models, layers, and images. Color is **aligned** across all models and layers.')
        gr.Markdown('')
        gr.Markdown("To see a good pattern, you will need to load 100~1000 images. 100 images need 10sec for RTX4090. Running out of HuggingFace GPU Quota? Try [Demo](https://ncut-pytorch.readthedocs.io/en/latest/demo/) hosted at UPenn")
        gr.Markdown('---')
        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()

                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section(advanced=True, is_random=True)
                num_images_slider.value = 100
                
                
            with gr.Column(scale=5, min_width=200):
                output_gallery = make_output_images_section()
                gr.Markdown('### TIP1: use the `full-screen` button, and use `arrow keys` to navigate')
                gr.Markdown('---')
                gr.Markdown('Model: CLIP(ViT-B-16/openai), DiNOv2reg(dinov2_vitb14_reg), MAE(vit_base)')
                gr.Markdown('Layer type: attention output (attn), without sum of residual')
                gr.Markdown('### TIP2: for large image set, please increase the `num_sample` for t-SNE and NCUT')
                gr.Markdown('---')
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                model_dropdown.value = "AlignedThreeModelAttnNodes"
                model_dropdown.visible = False
                layer_slider.visible = False
                node_type_dropdown.visible = False
                num_sample_ncut_slider.value = 10000
                num_sample_tsne_slider.value = 1000
                # logging text box
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
                
        # galleries = []
        # for i_model, model_name in enumerate(["CLIP", "DINO", "MAE"]):
        #     with gr.Row():
        #         for i_layer in range(1, 13):
        #             with gr.Column(scale=5, min_width=200):
        #                 gr.Markdown(f'### {model_name} Layer {i_layer}')
        #                 output_gallery = gr.Gallery(value=[], label="NCUT Embedding", show_label=False, elem_id="ncut", columns=[3], rows=[1], object_fit="contain", height="auto", show_fullscreen_button=True)
        #                 galleries.append(output_gallery)
                

        # clear_images_button.click(lambda x: [] * (len(galleries) + 1), outputs=[input_gallery] + galleries)
        clear_images_button.click(lambda x: ([], []), outputs=[input_gallery, output_gallery])
        
        false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
        no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
        
        submit_button.click(
            run_fn,
            inputs=[
                input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                positive_prompt, negative_prompt,
                false_placeholder, no_prompt, no_prompt, no_prompt,
                affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown
            ],
            # outputs=galleries + [logging_text],
            outputs=[output_gallery, logging_text],
        )
    
    with gr.Tab('Compare Models'): 
        def add_one_model(i_model=1):
            with gr.Column(scale=5, min_width=200) as col:
                gr.Markdown(f'### Output Images')
                output_gallery = gr.Gallery(format='png', value=[], label="NCUT Embedding", show_label=False, elem_id=f"ncut{i_model}", columns=[3], rows=[1], object_fit="contain", height="auto", show_fullscreen_button=True)
                submit_button = gr.Button("🔴 RUN", elem_id=f"submit_button{i_model}", variant='primary')
                add_output_images_buttons(output_gallery)
                [
                    model_dropdown, layer_slider, node_type_dropdown, num_eig_slider, 
                    affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                    embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                    perplexity_slider, n_neighbors_slider, min_dist_slider, 
                    sampling_method_dropdown, positive_prompt, negative_prompt
                ] = make_parameters_section()
                # logging text box
                logging_text = gr.Textbox("Logging information", label="Logging", elem_id="logging", type="text", placeholder="Logging information")
                false_placeholder = gr.Checkbox(label="False", value=False, elem_id="false_placeholder", visible=False)
                no_prompt = gr.Textbox("", label="", elem_id="empty_placeholder", type="text", placeholder="", visible=False)
                
                submit_button.click(
                    run_fn,
                    inputs=[
                        input_gallery, model_dropdown, layer_slider, num_eig_slider, node_type_dropdown, 
                        positive_prompt, negative_prompt,
                        false_placeholder, no_prompt, no_prompt, no_prompt,
                        affinity_focal_gamma_slider, num_sample_ncut_slider, knn_ncut_slider, 
                        embedding_method_dropdown, embedding_metric_dropdown, num_sample_tsne_slider, knn_tsne_slider, 
                        perplexity_slider, n_neighbors_slider, min_dist_slider, sampling_method_dropdown
                    ],
                    outputs=[output_gallery, logging_text]
                )
                
                return col

        with gr.Row():
            with gr.Column(scale=5, min_width=200):
                input_gallery, submit_button, clear_images_button = make_input_images_section()
                clear_images_button.click(lambda x: ([],), outputs=[input_gallery])
                submit_button.visible = False
                dataset_dropdown, num_images_slider, random_seed_slider, load_images_button = make_dataset_images_section(advanced=True)

                
            for i in range(2):
                add_one_model()
                
        # Create rows and buttons in a loop
        rows = []
        buttons = []

        for i in range(4):
            row = gr.Row(visible=False)
            rows.append(row)
            
            with row:
                for j in range(3):
                    with gr.Column(scale=5, min_width=200):
                        add_one_model()

            button = gr.Button("➕ Add Compare", elem_id=f"add_button_{i}", visible=False if i > 0 else True, scale=3)
            buttons.append(button)
            
            if i > 0:
                # Reveal the current row and next button
                buttons[i - 1].click(fn=lambda x: gr.update(visible=True), outputs=row)
                buttons[i - 1].click(fn=lambda x: gr.update(visible=True), outputs=button)
                
                # Hide the current button
                buttons[i - 1].click(fn=lambda x: gr.update(visible=False), outputs=buttons[i - 1])

        # Last button only reveals the last row and hides itself
        buttons[-1].click(fn=lambda x: gr.update(visible=True), outputs=rows[-1])
        buttons[-1].click(fn=lambda x: gr.update(visible=False), outputs=buttons[-1])
    
    
    with gr.Tab('📄About'):
        gr.Markdown("**This demo is for the Python package `ncut-pytorch`, please visit the [Documentation](https://ncut-pytorch.readthedocs.io/)**")
        gr.Markdown("**All the models and functions used for this demo are in the Python package `ncut-pytorch`**")
        gr.Markdown("---")
        gr.Markdown("---")
        gr.Markdown("**Normalized Cuts**, aka. spectral clustering, is a graphical method to analyze data grouping in the affinity eigenvector space. It has been widely used for unsupervised segmentation in the 2000s.")
        gr.Markdown("*Normalized Cuts and Image Segmentation, Jianbo Shi and Jitendra Malik, 2000*")
        gr.Markdown("---")        
        gr.Markdown("**We have improved NCut, with some advanced features:**")
        gr.Markdown("- **Nyström** Normalized Cut, is a new approximation algorithm developed for large-scale graph cuts, a large-graph of million nodes can be processed in under 10s (cpu) or 2s (gpu).")
        gr.Markdown("- **spectral-tSNE** visualization, a new method to visualize the high-dimensional eigenvector space with 3D RGB cube. Color is aligned across images, color infers distance in representation.")
        gr.Markdown("*paper in prep, Yang 2024*")
        gr.Markdown("*AlignedCut: Visual Concepts Discovery on Brain-Guided Universal Feature Space, Huzheng Yang, James Gee\*, and Jianbo Shi\*, 2024*")
        gr.Markdown("---")        
        gr.Markdown("---")        
        gr.Markdown('<p style="text-align: center;">We thank the HuggingFace team for hosting this demo.</p>')        
        
        
    with gr.Row():
        with gr.Column():
            gr.Markdown("##### This demo is for `ncut-pytorch`, [Documentation](https://ncut-pytorch.readthedocs.io/) ")
        with gr.Column():
            gr.Markdown("###### Running out of GPU? Try [Demo](https://ncut-pytorch.readthedocs.io/en/latest/demo/) hosted at UPenn")

# for local development
if os.path.exists("/hf_token.txt"):
    os.environ["HF_ACCESS_TOKEN"] = open("/hf_token.txt").read().strip()

if DOWNLOAD_ALL_MODELS_DATASETS:
    from ncut_pytorch.backbone import download_all_models
    threading.Thread(target=download_all_models).start()
    from ncut_pytorch.backbone_text import download_all_models
    threading.Thread(target=download_all_models).start()
    
    threading.Thread(target=download_all_datasets).start()
    
demo.launch(share=True)


# # %%
# # debug
# # change working directory to "/"
# os.chdir("/")
# images = [(Image.open(image), None) for image in default_images]
# ret = run_fn(images, num_eig=30)
# # %%