File size: 11,595 Bytes
51fedf6
 
 
 
 
 
 
 
 
c4fbd39
51fedf6
 
 
 
 
6df49f3
51fedf6
 
 
c4fbd39
51fedf6
6df49f3
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a10ced8
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
a10ced8
51fedf6
 
 
 
 
6df49f3
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6df49f3
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c4fbd39
 
 
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
c4fbd39
51fedf6
6df49f3
51fedf6
 
 
 
 
 
 
c4fbd39
 
 
 
 
 
 
 
 
 
 
 
 
 
51fedf6
 
 
 
 
 
 
 
 
 
6df49f3
 
51fedf6
 
 
 
 
 
 
 
 
 
 
 
6df49f3
 
51fedf6
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
6df49f3
51fedf6
 
 
 
6df49f3
51fedf6
 
 
 
 
 
 
 
 
 
 
c4fbd39
 
51fedf6
6df49f3
da8234d
6df49f3
51fedf6
c4fbd39
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
import os
import torch
import numpy as np
import lightning.pytorch as pl
import gradio as gr
import imageio
import random
import matplotlib.pyplot as plt
import cv2
import skdim

from torch.utils.data import Dataset, DataLoader

from PIL import Image
from matplotlib import cm
from safetensors.torch import save_file, load_file
from sklearn.cluster import AgglomerativeClustering
from sklearn.manifold import TSNE
from sklearn.neighbors import KDTree
from sklearn.preprocessing import StandardScaler

from minimal_script import EmbeddingNetwork, closest_interval, adj_size, PLModule


class PredictDataset(Dataset):
    def __init__(self, data_dir, sample=None):
        self.image_paths = []
        extensions = ('jpg', 'jpeg', 'png', 'tif', 'webp')
        for fname in sorted(os.listdir(data_dir)):
            if any(fname.lower().endswith(ext) for ext in extensions):
                self.image_paths.append(os.path.join(data_dir, fname))
        if sample:
            self.image_paths = random.sample(self.image_paths, sample)

    def __len__(self):
        return len(self.image_paths)

    def __getitem__(self, idx):
        path = self.image_paths[idx]
        image = imageio.v3.imread(path).copy()
        image = torch.from_numpy(image).permute(2, 0, 1)
        processed = closest_interval(adj_size(image, 1024))
        processed = 2*(processed/255)-1
        return processed.detach(), path


def explore_embedding_space(embeddings, image_paths, model):
    """
    Create an interface for exploring N-dimensional image embeddings

    Args:
        embeddings: NumPy array of shape [B, N]
        image_paths: List of B image file paths
    """
    # Validate inputs
    assert len(embeddings) == len(image_paths), "Mismatch between embeddings and image paths"
    assert embeddings.ndim == 2, "Embeddings should be 2-dimensional"

    # Precompute min/max for each dimension
    min_vals = embeddings.min(axis=0)
    max_vals = embeddings.max(axis=0)
    ranges = max_vals - min_vals

    # Build KDTree for efficient nearest neighbor search
    tree = KDTree(embeddings)

    # Create initial point (mean of embeddings)
    initial_point = embeddings.mean(axis=0).tolist()

    # Create slider components for each dimension
    sliders = []
    for i in range(embeddings.shape[1]):
        slider = gr.Slider(
            float(min_vals[i]),
            float(max_vals[i]),
            value=float(initial_point[i]),
            step=float(ranges[i]) / 100,
            label=f"Dimension {i + 1}"
        )
        sliders.append(slider)

    def compute_gradient_heatmap(image_path):
        """Compute gradient heatmap for an image"""
        # Load and preprocess image
        img = imageio.v3.imread(image_path).copy()
        img = torch.from_numpy(img).permute(2, 0, 1)
        img_tensor = closest_interval(adj_size(img, 1024)).unsqueeze(0)
        img_tensor = 2*(img_tensor/255)-1
        img_tensor.requires_grad_(True)

        # Move to GPU if available
        device = 'cuda' if torch.cuda.is_available() else 'cpu'
        img_tensor = img_tensor.to(device).to(torch.float16)

        # Compute embedding and gradient
        with torch.enable_grad():
            embd = model(img_tensor)
            norm = embd.norm(p=2, dim=1).sum()
            grad = torch.autograd.grad(norm, img_tensor, retain_graph=False)[0]

        # Compute gradient magnitude
        grad_mag = grad.squeeze(0).norm(dim=0).detach().cpu().numpy()

        # Normalize and apply colormap
        grad_min, grad_max = grad_mag.min(), grad_mag.max()
        if grad_max > grad_min:
            grad_norm = (grad_mag - grad_min) / (grad_max - grad_min)
        else:
            grad_norm = grad_mag * 0  # Handle uniform case

        heatmap = cm.jet(grad_norm)[..., :3]  # Use jet colormap
        return heatmap

    def overlay_heatmap(original_img, heatmap, alpha=0.4):
        """Overlay heatmap on original image"""
        # Resize heatmap to match original image
        heatmap_img = Image.fromarray((heatmap * 255).astype(np.uint8))
        heatmap_img = heatmap_img.resize(original_img.size)

        # Convert original to RGBA and heatmap to RGBA
        #original_rgba = original_img.convert("RGBA")
        #heatmap_rgba = heatmap_img.convert("RGBA")

        # Blend images
        blended = Image.blend(original_img, heatmap_img, alpha)
        return blended

    def get_overlay_image(image_path):
        """Get image with gradient overlay"""
        img = Image.open(image_path).convert('RGB')
        #heatmap = compute_gradient_heatmap(image_path)
        #return overlay_heatmap(img, heatmap)
        return img

    def add_caption_to_image(image, caption):
        """Add text caption to the bottom of an image"""
        # Convert to OpenCV format
        if isinstance(image, Image.Image):
            img = np.array(image)
        else:
            img = image.copy()

        # Add black bar at bottom
        bar_height = 30
        img = cv2.copyMakeBorder(img, 0, bar_height, 0, 0, cv2.BORDER_CONSTANT, value=[0, 0, 0])

        # Add white text
        font = cv2.FONT_HERSHEY_SIMPLEX
        text_size = cv2.getTextSize(caption, font, 0.5, 1)[0]
        text_x = (img.shape[1] - text_size[0]) // 2
        text_y = img.shape[0] - 10
        cv2.putText(img, caption, (text_x, text_y), font, 0.5, (255, 255, 255), 1)

        return Image.fromarray(img)

    # Function to find nearby images
    def find_nearby_images(*point):
        point = np.array(point).reshape(1, -1)
        distances, indices = tree.query(point, k=8)
        indices = indices[0]
        distances = distances[0]

        # Get paths and create overlay images
        paths = [image_paths[i] for i in indices]
        images_with_gradients = [get_overlay_image(p) for p in paths]

        # Create images with baked-in captions
        final_images = []
        for img, dist in zip(images_with_gradients, distances):
            caption = f"Dist: {dist:.2f}"
            final_img = add_caption_to_image(img, caption)
            final_images.append(final_img)

        warning = ""
        if distances[0] > 5.0:  # Warn if nearest image is far
            warning = "⚠️ Nearest image is far (distance={:.2f}). Consider adjusting sliders.".format(distances[0])

        return final_images, warning

    # Build interface
    with gr.Blocks() as demo:
        gr.Markdown("## N-Dimensional Embedding Space Explorer")
        gr.Markdown("Adjust sliders to navigate. Images show gradient of embedding norm w.r.t. input.")

        # Warning output
        warning = gr.Textbox(label="Status", interactive=False)

        # Gallery for images
        gallery = gr.Gallery(
            label="Nearest Images (Distance Ordered)",
            columns=4,
            object_fit="contain",
            height="auto",
            show_label=True,
        )

        # Create sliders in a compact row
        with gr.Row():
            for slider in sliders:
                slider.render()

        # Connect slider changes to update function
        for slider in sliders:
            slider.change(
                find_nearby_images,
                inputs=sliders,
                outputs=[gallery, warning]
            )

        # Initial trigger
        demo.load(
            find_nearby_images,
            inputs=sliders,
            outputs=[gallery, warning]
        )

    return demo



def generate_embeddings(image_folder, mode, model):
    predict_dataset = PredictDataset(image_folder, 5000)
    predict_loader = DataLoader(predict_dataset, batch_size=1, num_workers=5, pin_memory=True)
    trainer = pl.Trainer(accelerator="gpu", logger=False, enable_checkpointing=False, precision="16-mixed")
    predictions_0 = trainer.predict(model, predict_loader)
    predictions = torch.cat([pred[0] for pred in predictions_0], dim=0).numpy()
    paths = []
    for pred in predictions_0:
        for i in pred[1]:
            paths.append(i)
    if mode == 'Grouping':
        #estimate global intrinsic dimension
        #scaler = StandardScaler()
        #normalised_predictions = scaler.fit_transform(predictions)
        # Initialize estimators
        estimators = [skdim.id.TwoNN(), skdim.id.CorrInt(), skdim.id.DANCo()]
        results = {}
        
        for est in estimators:
            est.fit(predictions)
            results[type(est).__name__] = est.dimension_

        print("Intrinsic Dimension Estimates:")
        for name, dim in results.items():
            print(f"{name}: {dim:.2f}")
        labels = cluster_embeddings(predictions)

        row_norms = np.linalg.norm(predictions, axis=1)
        average_norms = np.mean(np.abs(predictions), axis=0)
        plt.figure(figsize=(8, 5))
        plt.bar(range(predictions.shape[1]), average_norms, color='skyblue')
        plt.xlabel('Feature Index (C)')
        plt.ylabel('Average Norm')
        plt.title(f'Average Norm for Each Feature (Column)')
        plt.xticks(range(predictions.shape[1]))
        #plt.show()
        plt.savefig('Norms.png')

        plt.figure(figsize=(8, 6))
        tsne = TSNE(n_components=2, random_state=42)
        reduced_data = tsne.fit_transform(predictions)
        plt.scatter(reduced_data[:, 0], reduced_data[:, 1], c=row_norms, cmap='viridis', s=50, edgecolor='k', label="Data Points")
        plt.colorbar(label='Norm Value')
        plt.xlabel('Feature 1')
        plt.ylabel('Feature 2')
        plt.title(f'Scatter Plot of Data Points and Average Norm')
        plt.legend()
        plt.grid(True)
        plt.axis('equal')
        #plt.show()
        plt.savefig('Groups.png')

        # List unique clusters
        unique_clusters = np.unique(labels)
        # Gradio UI
        with gr.Blocks() as demo:
            gr.Markdown("## Explore Image Clusters by Style")

            # Dropdown for selecting a cluster
            cluster_selector = gr.Dropdown(choices=unique_clusters.tolist(), label="Select Cluster to Explore")

            # Gallery to display images
            image_gallery = gr.Gallery(label="Sample Images from Selected Cluster")


            # Gradio Interface for Cluster Exploration
            def explore_clusters(cluster_idx):
                # Find images that belong to the selected cluster
                cluster_images = [paths[i] for i in range(len(labels)) if labels[i] == cluster_idx]
                # Load and return images
                images = [Image.open(img_path) for img_path in cluster_images[:50]]  # Show a sample of 50 images
                return images

            # Update function for the gallery
            cluster_selector.change(fn=explore_clusters, inputs=cluster_selector, outputs=image_gallery)

        demo.launch()
    elif mode == 'Explore':
        demo = explore_embedding_space(predictions, paths, model.to('cuda').to(torch.float16))
        demo.launch()


# Apply Agglomerative Clustering
def cluster_embeddings(predictions, distance_threshold=32.0):
    agg_clustering = AgglomerativeClustering(
        n_clusters=None,
        distance_threshold=distance_threshold,
        linkage='ward'
    )
    labels = agg_clustering.fit_predict(predictions)
    return labels



if __name__ == '__main__':
    folder = 'Enter Images folder name here'
    #folder = 'images_for_style_embedding'
    device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
    model = PLModule()
    state_dict = load_file("Style_Embedder_v3.safetensors")
    model.network.load_state_dict(state_dict)
    # 'Grouping' or 'Explore'
    generate_embeddings(folder, 'Grouping', model)