File size: 14,652 Bytes
af6e161
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
import numpy as np
import random
import torch
import torch.nn as nn
from collections import defaultdict
import cv2
import pickle
import os
from PIL import Image


class Config:
    """Generation system configuration"""

    IMG_SIZE = 128
    PATCH_SIZE = 16
    GRID_SIZE = 8
    PATTERN_DIM = 32

    PATTERN_TYPES = ["sobel", "haar", "gradient", "texture", "color", "frequency"]

    @property
    def active_pattern_types(self):
        return self.PATTERN_TYPES

    @property
    def n_active_patterns(self):
        return len(self.active_pattern_types)


class PatternExtractor:
    """Extract various pattern types from images"""

    def __init__(self):
        self.pattern_dim = Config.PATTERN_DIM

    def sobel_filter(self, patch):
        """Sobel filters for edge detection"""
        gray = torch.mean(patch, dim=0).numpy()
        sobel_x = cv2.Sobel(gray, cv2.CV_64F, 1, 0, ksize=3)
        sobel_y = cv2.Sobel(gray, cv2.CV_64F, 0, 1, ksize=3)

        sobel_x = self._normalize_and_reduce(sobel_x)
        sobel_y = self._normalize_and_reduce(sobel_y)

        return np.concatenate([sobel_x, sobel_y])

    def haar_features(self, patch):
        """Haar-like features"""
        gray = torch.mean(patch, dim=0).numpy()
        h, w = gray.shape

        left_half = gray[:, : w // 2]
        right_half = gray[:, w // 2 :]
        haar_vertical = np.mean(left_half) - np.mean(right_half)

        top_half = gray[: h // 2, :]
        bottom_half = gray[h // 2 :, :]
        haar_horizontal = np.mean(top_half) - np.mean(bottom_half)

        haar_features = np.array([haar_vertical, haar_horizontal])
        return self._expand_to_dim(haar_features)

    def gradient_features(self, patch):
        """Gradient features"""
        gradients = []
        for c in range(patch.shape[0]):
            channel = patch[c].numpy()
            grad_x = np.gradient(channel, axis=1)
            grad_y = np.gradient(channel, axis=0)

            grad_stats = np.array(
                [
                    np.mean(np.abs(grad_x)),
                    np.mean(np.abs(grad_y)),
                    np.std(grad_x),
                    np.std(grad_y),
                ]
            )
            gradients.extend(grad_stats)

        return self._expand_to_dim(np.array(gradients))

    def texture_features(self, patch):
        """Texture features"""
        gray = torch.mean(patch, dim=0).numpy()
        h, w = gray.shape
        lbp_features = []

        for i in range(1, h - 1):
            for j in range(1, w - 1):
                center = gray[i, j]
                pattern = 0
                for di in [-1, 0, 1]:
                    for dj in [-1, 0, 1]:
                        if di == 0 and dj == 0:
                            continue
                        if gray[i + di, j + dj] > center:
                            pattern += 1
                lbp_features.append(pattern)

        lbp_stats = np.array(
            [
                np.mean(lbp_features),
                np.std(lbp_features),
                np.max(lbp_features),
                np.min(lbp_features),
            ]
        )

        return self._expand_to_dim(lbp_stats)

    def color_features(self, patch):
        """Color features"""
        color_stats = []
        for c in range(patch.shape[0]):
            channel = patch[c]
            stats = torch.tensor(
                [
                    torch.mean(channel),
                    torch.std(channel),
                    torch.max(channel),
                    torch.min(channel),
                ]
            )
            color_stats.extend(stats.numpy())

        return self._expand_to_dim(np.array(color_stats))

    def frequency_features(self, patch):
        """Frequency features via FFT"""
        gray = torch.mean(patch, dim=0).numpy()
        fft = np.fft.fft2(gray)
        fft_magnitude = np.abs(fft)

        freq_stats = np.array(
            [
                np.mean(fft_magnitude),
                np.std(fft_magnitude),
                np.sum(fft_magnitude[: len(fft_magnitude) // 4, :]),
                np.sum(fft_magnitude[3 * len(fft_magnitude) // 4 :, :]),
            ]
        )

        return self._expand_to_dim(freq_stats)

    def _normalize_and_reduce(self, array):
        """Normalize and reduce to smaller dimension"""
        array = (array - np.min(array)) / (np.max(array) - np.min(array) + 1e-8)
        return array.flatten()[: self.pattern_dim // 2]

    def _expand_to_dim(self, features):
        """Expand features to required dimension"""
        if len(features) >= self.pattern_dim:
            return features[: self.pattern_dim]
        else:
            expanded = np.zeros(self.pattern_dim)
            expanded[: len(features)] = features
            return expanded

    def extract_all_patterns(self, patch):
        """Extract all pattern types"""
        patterns = {}
        patterns["sobel"] = self.sobel_filter(patch)
        patterns["haar"] = self.haar_features(patch)
        patterns["gradient"] = self.gradient_features(patch)
        patterns["texture"] = self.texture_features(patch)
        patterns["color"] = self.color_features(patch)
        patterns["frequency"] = self.frequency_features(patch)

        return patterns


class PatternReconstructor(nn.Module):
    """Neural network for reconstructing RGB patches from patterns"""

    def __init__(self, config):
        super(PatternReconstructor, self).__init__()

        input_dim = config.n_active_patterns * config.PATTERN_DIM
        hidden_dim = 512

        self.encoder = nn.Sequential(
            nn.Linear(input_dim, hidden_dim),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(hidden_dim, hidden_dim // 2),
            nn.ReLU(),
            nn.Dropout(0.2),
            nn.Linear(hidden_dim // 2, hidden_dim // 4),
            nn.ReLU(),
        )

        self.decoder_conv = nn.Sequential(
            nn.ConvTranspose2d(128, 128, 4, 1, 0),
            nn.ReLU(),
            nn.ConvTranspose2d(128, 64, 4, 2, 1),
            nn.ReLU(),
            nn.ConvTranspose2d(64, 32, 4, 2, 1),
            nn.ReLU(),
            nn.ConvTranspose2d(32, 3, 3, 1, 1),
            nn.Tanh(),
        )

    def forward(self, patterns):
        encoded = self.encoder(patterns)
        encoded = encoded.view(-1, 128, 1, 1)
        reconstructed = self.decoder_conv(encoded)
        return reconstructed


class MultiPatternMarkovChain:
    """Multi-dimensional Markov chain for patterns"""

    def __init__(self, config):
        self.config = config
        self.chains = {}
        self.pattern_store = {}

        for pattern_type in config.active_pattern_types:
            self.chains[pattern_type] = defaultdict(lambda: defaultdict(list))

    def generate_patterns(self, grid_size, temperature=1.0):
        """Generate patterns for all active types"""
        result = {}

        for pattern_type in self.config.active_pattern_types:
            if pattern_type in self.chains:
                result[pattern_type] = self._generate_single_pattern_grid(
                    pattern_type, grid_size, temperature
                )

        return result

    def _generate_single_pattern_grid(self, pattern_type, grid_size, temperature):
        """Generate grid for single pattern type"""
        chain = self.chains[pattern_type]
        grid = [[None for _ in range(grid_size)] for _ in range(grid_size)]

        if not chain:
            return grid

        start_hash = random.choice(list(chain.keys()))
        grid[0][0] = start_hash

        for i in range(grid_size):
            for j in range(grid_size):
                if i == 0 and j == 0:
                    continue

                candidates = []

                if i > 0 and grid[i - 1][j] is not None:
                    candidates.extend(chain[grid[i - 1][j]].get("down", []))

                if j > 0 and grid[i][j - 1] is not None:
                    candidates.extend(chain[grid[i][j - 1]].get("right", []))

                if candidates:
                    grid[i][j] = self._softmax_selection(candidates, temperature)
                else:
                    grid[i][j] = random.choice(list(chain.keys()))

        return grid

    def _softmax_selection(self, candidates, temperature):
        """Selection with temperature consideration"""
        if not candidates:
            return None

        if len(candidates) == 1:
            return candidates[0][0]

        probs = np.array([c[1] for c in candidates])

        if temperature == 0:
            max_idx = np.argmax(probs)
            return candidates[max_idx][0]
        else:
            probs = probs / temperature
            exp_probs = np.exp(probs - np.max(probs))
            softmax_probs = exp_probs / np.sum(exp_probs)

            idx = np.random.choice(len(candidates), p=softmax_probs)
            return candidates[idx][0]


class VisualMarkovGenerator:
    """Main image generator class"""

    def __init__(

        self,

        model_path="pattern_reconstructor_final.pth",

        markov_path="markov_chains.pkl",

    ):
        self.config = Config()
        self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
        self.model = PatternReconstructor(self.config).to(self.device)

        if os.path.exists(model_path):
            self.model.load_state_dict(torch.load(model_path, map_location=self.device))
            print(f"Model loaded from {model_path}")
        else:
            print(f"Model file {model_path} not found. Using random weights.")

        if os.path.exists(markov_path):
            with open(markov_path, "rb") as f:
                save_data = pickle.load(f)

            self.markov_chain = MultiPatternMarkovChain(self.config)
            self.markov_chain.pattern_store = save_data["pattern_store"]

            for pattern_type in self.config.active_pattern_types:
                if pattern_type in save_data["chains"]:
                    chain_data = save_data["chains"][pattern_type]
                    for state, directions in chain_data.items():
                        for direction, candidates in directions.items():
                            self.markov_chain.chains[pattern_type][state][
                                direction
                            ] = candidates

            print(f"Markov chains loaded from {markov_path}")
        else:
            print(f"Markov chains file {markov_path} not found. Creating empty chain.")
            self.markov_chain = MultiPatternMarkovChain(self.config)

        self.model.eval()

    def generate_image(self, temperature=1.0, seed=None):
        """Generate image with specified temperature"""
        if seed is not None:
            random.seed(seed)
            np.random.seed(seed)
            torch.manual_seed(seed)

        pattern_grids = self.markov_chain.generate_patterns(
            self.config.GRID_SIZE, temperature=temperature
        )

        if not pattern_grids:
            print("Warning: empty patterns for generation")
            return self._create_random_image()

        result_image = self._generate_image_from_patterns(pattern_grids)

        img_np = torch.clamp(result_image, 0, 1).permute(1, 2, 0).numpy()
        img_np = (img_np * 255).astype(np.uint8)

        return Image.fromarray(img_np)

    def _generate_image_from_patterns(self, pattern_grids):
        """Generate image from patterns"""
        first_pattern_type = list(pattern_grids.keys())[0]
        grid_size = len(pattern_grids[first_pattern_type])
        result_image = torch.zeros(
            (3, grid_size * self.config.PATCH_SIZE, grid_size * self.config.PATCH_SIZE)
        )

        with torch.no_grad():
            for i in range(grid_size):
                for j in range(grid_size):
                    pattern_vector = []

                    for pattern_type in self.config.active_pattern_types:
                        if pattern_type in pattern_grids:
                            pattern_hash = pattern_grids[pattern_type][i][j]
                            if pattern_hash in self.markov_chain.pattern_store:
                                pattern = self.markov_chain.pattern_store[pattern_hash]
                                pattern_vector.extend(pattern)
                            else:
                                pattern_vector.extend(np.zeros(self.config.PATTERN_DIM))
                        else:
                            pattern_vector.extend(np.zeros(self.config.PATTERN_DIM))

                    if len(pattern_vector) > 0:
                        pattern_tensor = (
                            torch.FloatTensor(pattern_vector)
                            .unsqueeze(0)
                            .to(self.device)
                        )
                        reconstructed_patch = (
                            self.model(pattern_tensor).cpu().squeeze(0)
                        )

                        start_i = i * self.config.PATCH_SIZE
                        end_i = start_i + self.config.PATCH_SIZE
                        start_j = j * self.config.PATCH_SIZE
                        end_j = start_j + self.config.PATCH_SIZE

                        result_image[:, start_i:end_i, start_j:end_j] = (
                            reconstructed_patch
                        )

        return result_image

    def _create_random_image(self):
        """Create random image in case of error"""
        random_img = torch.rand(3, self.config.IMG_SIZE, self.config.IMG_SIZE)
        img_np = (random_img.permute(1, 2, 0).numpy() * 255).astype(np.uint8)
        return Image.fromarray(img_np)


def create_generator(

    model_path="pattern_reconstructor_final.pth", markov_path="markov_chains.pkl"

):
    """Create generator instance"""
    return VisualMarkovGenerator(model_path, markov_path)


if __name__ == "__main__":
    generator = create_generator()
    stats = generator.get_stats()
    print("Generator statistics:")
    print(f"Total patterns: {stats['total_patterns']}")
    print(f"Model loaded: {stats['model_loaded']}")

    img = generator.generate_image(temperature=1.0, seed=42)
    print(f"Generated image size: {img.size}")