File size: 8,007 Bytes
be305fb
 
 
 
 
 
ee07cd3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be305fb
 
 
ee07cd3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be305fb
 
 
 
 
ee07cd3
be305fb
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ee07cd3
 
be305fb
 
ee07cd3
be305fb
 
 
ee07cd3
 
 
 
be305fb
 
ee07cd3
 
 
be305fb
ee07cd3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be305fb
ee07cd3
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
be305fb
ee07cd3
be305fb
ee07cd3
 
 
 
 
be305fb
 
 
ee07cd3
be305fb
 
ee07cd3
 
 
 
 
 
be305fb
 
 
 
 
ee07cd3
 
be305fb
 
 
 
 
 
 
 
 
 
 
 
ee07cd3
be305fb
ee07cd3
 
 
 
be305fb
ee07cd3
 
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
import os
import numpy as np
from datasets import load_dataset
from PIL import Image, ImageOps, ImageFilter
from tqdm import tqdm
import random
import requests
import io
import time

def download_image(url, timeout=10, retries=2):
    """Download image from URL with retry mechanism"""
    for attempt in range(retries):
        try:
            headers = {
                'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36'
            }
            response = requests.get(url, timeout=timeout, headers=headers)
            
            if response.status_code == 200:
                image = Image.open(io.BytesIO(response.content))
                return image
            else:
                return None
                
        except Exception as e:
            if attempt == retries - 1:  # Last attempt
                print(f"Failed to download {url}: {e}")
                return None
            time.sleep(0.5)  # Brief pause before retry
    
    return None

def preprocess_image(image, target_size=512, quality_threshold=0.7):
    """Preprocess image with various enhancements"""
    if image is None:
        return None
        
    try:
        # Convert to RGB if needed
        if image.mode != 'RGB':
            image = image.convert('RGB')
        
        # Filter out low quality images
        width, height = image.size
        if min(width, height) < target_size * quality_threshold:
            return None
        
        # Center crop to square if not already
        if width != height:
            size = min(width, height)
            left = (width - size) // 2
            top = (height - size) // 2
            image = image.crop((left, top, left + size, top + size))
        
        # Resize to target size
        image = image.resize((target_size, target_size), Image.Resampling.LANCZOS)
        
        # Enhance image quality
        # Slightly sharpen
        image = image.filter(ImageFilter.UnsharpMask(radius=0.5, percent=120, threshold=3))
        
        # Auto-adjust levels
        image = ImageOps.autocontrast(image, cutoff=1)
        
        return image
        
    except Exception as e:
        print(f"Error preprocessing image: {e}")
        return None

def clean_prompt(prompt):
    """Clean and normalize prompts"""
    if not prompt:
        return None
    
    # Remove excessive whitespace
    prompt = ' '.join(prompt.split())
    
    # Remove common artifacts
    prompt = prompt.replace('  ', ' ')
    prompt = prompt.strip(' .,;:')
    
    # Filter out very short or very long prompts
    words = prompt.split()
    if len(words) < 3 or len(words) > 50:
        return None
        
    return prompt

def prepare_dreambooth_data():
    # Load dataset
    print("Loading LAION dataset...")
    dataset = load_dataset("laion/laion2B-en-aesthetic", split="train", streaming=True)
    
    # Create directory structure
    data_dir = "./laion_dataset"
    os.makedirs(data_dir, exist_ok=True)
    
    valid_samples = 0
    processed_count = 0
    max_samples = 1000  # Limit total samples to process
    
    print(f"Starting to process up to {max_samples} samples...")
    
    # Process images with preprocessing
    for idx, sample in enumerate(tqdm(dataset, desc="Processing LAION samples")):
        if processed_count >= max_samples:
            break
            
        processed_count += 1
        
        try:
            # Get URL and text from LAION format
            image_url = sample.get('URL', '')
            text_prompt = sample.get('TEXT', '')
            
            if not image_url or not text_prompt:
                continue
            
            # Clean prompt first
            prompt = clean_prompt(text_prompt)
            if prompt is None:
                continue
            
            # Download image from URL
            print(f"Downloading image {valid_samples + 1}: {image_url[:50]}...")
            image = download_image(image_url)
            if image is None:
                continue
            
            # Preprocess downloaded image
            processed_image = preprocess_image(image)
            if processed_image is None:
                continue
            
            # Save processed image
            image_path = os.path.join(data_dir, f"image_{valid_samples:04d}.jpg")
            processed_image.save(image_path, "JPEG", quality=95, optimize=True)
            
            # Save cleaned caption
            caption_path = os.path.join(data_dir, f"image_{valid_samples:04d}.txt")
            with open(caption_path, 'w', encoding='utf-8') as f:
                f.write(prompt)
            
            valid_samples += 1
            
            # Optional: Add metadata file
            metadata_path = os.path.join(data_dir, f"image_{valid_samples-1:04d}_meta.txt")
            with open(metadata_path, 'w', encoding='utf-8') as f:
                f.write(f"URL: {image_url}\n")
                f.write(f"Aesthetic: {sample.get('aesthetic', 'N/A')}\n")
                f.write(f"Width: {sample.get('WIDTH', 'N/A')}\n")
                f.write(f"Height: {sample.get('HEIGHT', 'N/A')}\n")
            
            # Stop if we have enough samples
            if valid_samples >= 100:  # Adjust this number as needed
                break
                
        except Exception as e:
            print(f"Error processing sample {idx}: {e}")
            continue
    
    print(f"Processed {processed_count} samples, saved {valid_samples} valid images to {data_dir}")
    return data_dir

def create_demo_dataset():
    """Create demo dataset as last resort"""
    print("Creating demo dataset...")
    
    data_dir = "./demo_dataset"
    os.makedirs(data_dir, exist_ok=True)
    
    demo_prompts = [
        "a beautiful landscape with mountains",
        "portrait of a person with detailed features",
        "abstract colorful digital artwork",
        "modern architecture building design",
        "natural forest scene with trees",
        "urban cityscape at sunset",
        "artistic oil painting style",
        "vintage photography aesthetic",
        "minimalist geometric composition",
        "vibrant surreal art piece"
    ]
    
    for idx, prompt in enumerate(demo_prompts):
        # Create gradient background
        color1 = (random.randint(50, 200), random.randint(50, 200), random.randint(50, 200))
        color2 = (random.randint(100, 255), random.randint(100, 255), random.randint(100, 255))
        
        image = Image.new('RGB', (512, 512), color1)
        
        # Save files
        image_path = os.path.join(data_dir, f"image_{idx:04d}.jpg")
        image.save(image_path, "JPEG", quality=95)
        
        caption_path = os.path.join(data_dir, f"image_{idx:04d}.txt")
        with open(caption_path, 'w', encoding='utf-8') as f:
            f.write(prompt)
    
    print(f"Created {len(demo_prompts)} demo samples")
    return data_dir

# Main execution with fallback
def main():
    data_dir = prepare_dreambooth_data()
    
    # Generate training command
    training_command = f"""
accelerate launch \\
  --deepspeed_config_file ds_config.json \\
  diffusers/examples/dreambooth/train_dreambooth.py \\
    --pretrained_model_name_or_path="runwayml/stable-diffusion-v1-5" \\
    --instance_data_dir="{data_dir}" \\
    --instance_prompt="a high quality image" \\
    --output_dir="./laion-model" \\
    --resolution=512 \\
    --train_batch_size=1 \\
    --gradient_accumulation_steps=1 \\
    --gradient_checkpointing \\
    --learning_rate=5e-6 \\
    --lr_scheduler="constant" \\
    --lr_warmup_steps=0 \\
    --max_train_steps=400 \\
    --mixed_precision="fp16" \\
    --checkpointing_steps=100 \\
    --checkpoints_total_limit=1 \\
    --report_to="tensorboard" \\
    --logging_dir="./laion-model/logs"
"""
    
    print(f"\n✅ Dataset prepared in: {data_dir}")
    print("🚀 Run this command to train:")
    print(training_command)

if __name__ == "__main__":
    main()