File size: 8,253 Bytes
8462fc9
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import cv2
import numpy as np
import pytesseract
import torch
import torch.nn as nn

class NeuralNetworkDesigner:
    def __init__(self):
        self.layer_maps = {}

    def process_image(self, image_path):
        # Read the image
        image = cv2.imread(image_path)
        gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
        
        # Threshold the image
        _, binary = cv2.threshold(gray, 225, 255, cv2.THRESH_BINARY_INV)
        
        # Find contours
        contours, _ = cv2.findContours(binary, cv2.RETR_EXTERNAL, cv2.CHAIN_APPROX_SIMPLE)
        
        # Sort contours from top to bottom
        contours = sorted(contours, key=lambda c: cv2.boundingRect(c)[1])
        
        for i, contour in enumerate(contours):
            x, y, w, h = cv2.boundingRect(contour)
            roi = gray[y:y+h, x:x+w]
            
            # Perform OCR on the ROI
            text = pytesseract.image_to_string(roi).strip()
            self.parse_layer_info(i, text)

    def parse_layer_info(self, layer_index, text):
        lines = text.split('\n')
        layer_info = {'type': 'Unknown', 'text': text}
        
        try:
            if 'Input' in lines[0]:
                layer_info['type'] = 'Input'
                layer_info['channels'] = int(lines[1]) if len(lines) > 1 else None
            elif 'Conv' in lines[0]:
                layer_info['type'] = 'Conv2d'
                layer_info['out_channels'] = int(lines[1]) if len(lines) > 1 else None
                layer_info['kernel_size'] = int(lines[2]) if len(lines) > 2 else None
            elif any(x in lines[0] for x in ['MaxPool', 'AvgPool']):
                layer_info['type'] = 'MaxPool2d' if 'Max' in lines[0] else 'AvgPool2d'
                layer_info['kernel_size'] = int(lines[1]) if len(lines) > 1 else None
            elif 'Linear' in lines[0]:
                layer_info['type'] = 'Linear'
                if len(lines) > 1 and '*' in lines[1]:
                    layer_info['in_features'] = lines[1]
                layer_info['out_features'] = int(lines[-1]) if lines[-1].isdigit() else None
            elif 'BatchNorm' in lines[0]:
                layer_info['type'] = 'BatchNorm2d'
                layer_info['num_features'] = int(lines[1]) if len(lines) > 1 else None
            elif any(x in lines[0] for x in ['ReLU', 'LeakyReLU', 'Sigmoid', 'Tanh']):
                layer_info['type'] = lines[0]
            elif 'Dropout' in lines[0]:
                layer_info['type'] = 'Dropout'
                layer_info['p'] = float(lines[1]) if len(lines) > 1 else 0.5
            elif 'Transformer' in lines[0]:
                layer_info['type'] = 'Transformer'
                layer_info['d_model'] = int(lines[1]) if len(lines) > 1 else 512
                layer_info['nhead'] = int(lines[2]) if len(lines) > 2 else 8
            elif 'Attention' in lines[0]:
                layer_info['type'] = 'MultiheadAttention'
                layer_info['embed_dim'] = int(lines[1]) if len(lines) > 1 else 512
                layer_info['num_heads'] = int(lines[2]) if len(lines) > 2 else 8
            elif 'LSTM' in lines[0] or 'GRU' in lines[0]:
                layer_info['type'] = lines[0]
                layer_info['hidden_size'] = int(lines[1]) if len(lines) > 1 else 256
                layer_info['num_layers'] = int(lines[2]) if len(lines) > 2 else 1
        except ValueError as e:
            print(f"Error parsing layer {layer_index}: {e}")

        self.layer_maps[layer_index] = layer_info
        print(f"Parsed layer {layer_index}: {layer_info}")  # Debug print

    def generate_pytorch_code(self):
        code = "import torch\nimport torch.nn as nn\n\n"
        code += "class CustomNN(nn.Module):\n"
        code += "    def __init__(self):\n"
        code += "        super(CustomNN, self).__init__()\n"
        
        forward_code = "    def forward(self, x):\n"
        
        in_channels = None
        for i, layer_info in sorted(self.layer_maps.items()):
            if layer_info['type'] == 'Input':
                in_channels = layer_info.get('channels', 3)
                continue
            
            if layer_info['type'] == 'Conv2d':
                out_channels = layer_info.get('out_channels', 64)
                kernel_size = layer_info.get('kernel_size', 3)
                code += f"        self.conv{i} = nn.Conv2d({in_channels}, {out_channels}, kernel_size={kernel_size}, padding=1)\n"
                forward_code += f"        x = self.conv{i}(x)\n"
                in_channels = out_channels
            
            elif layer_info['type'] in ['MaxPool2d', 'AvgPool2d']:
                kernel_size = layer_info.get('kernel_size', 2)
                code += f"        self.pool{i} = nn.{layer_info['type']}(kernel_size={kernel_size})\n"
                forward_code += f"        x = self.pool{i}(x)\n"
            
            elif layer_info['type'] == 'Linear':
                out_features = layer_info.get('out_features')
                if i == 1 or (i > 1 and self.layer_maps[i-1]['type'] not in ['Linear', 'Flatten']):
                    code += f"        self.flatten = nn.Flatten()\n"
                    forward_code += f"        x = self.flatten(x)\n"
                    in_features = layer_info.get('in_features', 'x.shape[1]')
                else:
                    in_features = self.layer_maps[i-1].get('out_features', 64)
                code += f"        self.fc{i} = nn.Linear({in_features}, {out_features})\n"
                forward_code += f"        x = self.fc{i}(x)\n"
            
            elif layer_info['type'] == 'BatchNorm2d':
                num_features = layer_info.get('num_features', in_channels)
                code += f"        self.bn{i} = nn.BatchNorm2d({num_features})\n"
                forward_code += f"        x = self.bn{i}(x)\n"
            
            elif layer_info['type'] in ['ReLU', 'LeakyReLU', 'Sigmoid', 'Tanh']:
                code += f"        self.act{i} = nn.{layer_info['type']}()\n"
                forward_code += f"        x = self.act{i}(x)\n"
            
            elif layer_info['type'] == 'Dropout':
                p = layer_info.get('p', 0.5)
                code += f"        self.dropout{i} = nn.Dropout(p={p})\n"
                forward_code += f"        x = self.dropout{i}(x)\n"
            
            elif layer_info['type'] == 'Transformer':
                d_model = layer_info.get('d_model', 512)
                nhead = layer_info.get('nhead', 8)
                code += f"        self.transformer{i} = nn.Transformer(d_model={d_model}, nhead={nhead})\n"
                forward_code += f"        x = self.transformer{i}(x)\n"
            
            elif layer_info['type'] == 'MultiheadAttention':
                embed_dim = layer_info.get('embed_dim', 512)
                num_heads = layer_info.get('num_heads', 8)
                code += f"        self.attention{i} = nn.MultiheadAttention(embed_dim={embed_dim}, num_heads={num_heads})\n"
                forward_code += f"        x, _ = self.attention{i}(x, x, x)\n"
            
            elif layer_info['type'] in ['LSTM', 'GRU']:
                hidden_size = layer_info.get('hidden_size', 256)
                num_layers = layer_info.get('num_layers', 1)
                code += f"        self.rnn{i} = nn.{layer_info['type']}(input_size={in_channels}, hidden_size={hidden_size}, num_layers={num_layers}, batch_first=True)\n"
                forward_code += f"        x, _ = self.rnn{i}(x)\n"
            
            elif layer_info['type'] == 'Unknown':
                print(f"Warning: Unknown layer type at index {i}. Layer info: {layer_info}")
        
        code += "\n" + forward_code
        code += "        return x\n"
        
        return code

    def write_to_file(self, code, filename):
        with open(filename, 'w') as f:
            f.write(code)

    def design_network(self, image_path, output_file):
        self.process_image(image_path)
        pytorch_code = self.generate_pytorch_code()
        self.write_to_file(pytorch_code, output_file)
        # print(f"Neural network code has been generated and saved to '{output_file}'")
        # print("\nGenerated PyTorch Code:")
        # print(pytorch_code)