File size: 8,324 Bytes
f275c93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ffd0ad7
2d4137f
f275c93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
ed629f0
f275c93
 
 
 
 
 
 
ed629f0
f275c93
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
import argparse
from openai import OpenAI
import ast
import cairosvg
import json
import os
import utils
import traceback

from dotenv import load_dotenv
from PIL import Image
from prompts import sketch_first_prompt, system_prompt, gt_example


def call_argparse():
    parser = argparse.ArgumentParser(description='Process Arguments')
    
    # General
    parser.add_argument('--concept_to_draw', type=str, default="cat")
    parser.add_argument('--path2save', type=str, default=f"results/test")
    parser.add_argument('--temperature', type=float, default=0.3)
    parser.add_argument('--model', type=str, default='o3')
    parser.add_argument('--gen_mode', type=str, default='generation', choices=['generation', 'completion'])

    # Grid params
    parser.add_argument('--res', type=int, default=50, help="the resolution of the grid is set to 50x50")
    parser.add_argument('--cell_size', type=int, default=12, help="size of each cell in the grid")
    parser.add_argument('--stroke_width', type=float, default=7.0)

    args = parser.parse_args()
    args.grid_size = (args.res + 1) * args.cell_size

    args.save_name = args.concept_to_draw.replace(" ", "_")
    args.path2save = f"{args.path2save}/{args.save_name}"
    if not os.path.exists(args.path2save):
        os.makedirs(args.path2save)
        with open(f"{args.path2save}/experiment_log.json", 'w') as json_file:
            json.dump([], json_file, indent=4)
    return args


class SketchApp:
    def __init__(self, args):
        # General
        self.path2save = args.path2save
        self.target_concept = args.concept_to_draw

        # Grid related
        self.res = args.res
        self.num_cells = args.res
        self.cell_size = args.cell_size
        self.grid_size = (args.grid_size, args.grid_size)
        self.init_canvas, self.positions = utils.create_grid_image(res=args.res, cell_size=args.cell_size, header_size=args.cell_size)
        self.init_canvas_str = utils.image_to_str(self.init_canvas)
        self.cells_to_pixels_map = utils.cells_to_pixels(args.res, args.cell_size, header_size=args.cell_size)

        # SVG related 
        self.stroke_width = args.stroke_width
        
        # LLM Setup (you need to provide your OPENAI_API_KEY in your .env file)
        # self.cache = False
        self.max_tokens = 3000
        openai_key = os.getenv("OPENAI_API_KEY")
        self.client = OpenAI(api_key=openai_key)
        self.model = "gpt-4o"
        self.input_prompt = sketch_first_prompt.format(concept=args.concept_to_draw, gt_sketches_str=gt_example)
        self.gen_mode = args.gen_mode
        self.temperature = args.temperature
        

    def call_llm(self, system_message, other_msg, additional_args):
        response = self.client.chat.completions.create(
            model=self.model,
            messages=[{"role": "system", "content": system_message}] + other_msg,
            max_tokens=self.max_tokens,
            temperature=self.temperature,
            stop=additional_args.get("stop", None)
        )
        return response.choices[0].message.content

    
    def define_input_to_llm(self, msg_history, init_canvas_str, msg):
        content = []
        if init_canvas_str is not None:
            content.append({
                "type": "image_url",
                "image_url": "data:image/jpeg;base64," + init_canvas_str
            })
        content.append({"type": "text", "text": msg})

        other_msg = msg_history + [{"role": "user", "content": content}]
        return other_msg
        

    def get_response_from_llm(
        self,
        msg,
        system_message,
        msg_history=[],
        init_canvas_str=None,
        prefill_msg=None,
        # seed_mode="stochastic",
        stop=None,
        gen_mode="generation"
    ):  
        additional_args = {}
        # if seed_mode == "deterministic":
            #additional_args["temperature"] = 0.0
            #additional_args["top_k"] = 1

        # if self.cache:
        #     system_message = [{
        #         "type": "text",
        #         "text": system_message,
        #         "cache_control": {"type": "ephemeral"}
        #     }]

        # other_msg should contain all messgae without the system prompt
        other_msg = self.define_input_to_llm(msg_history, init_canvas_str, msg) 

        if gen_mode == "completion":
            if prefill_msg:
                other_msg = other_msg + [{"role": "assistant", "content": f"{prefill_msg}"}]
            
        # In case of stroke by stroke generation
        if stop:
            additional_args["stop"]= stop
        else:
            additional_args["stop"]= ["</answer>"]
 
        response = self.call_llm(system_message, other_msg, additional_args)
        content = response
        
        if gen_mode == "completion":
            other_msg = other_msg[:-1] # remove initial assistant prompt
            content = f"{prefill_msg}{content}" 

        # saves to json
        if self.path2save is not None:
            system_message_json = [{"role": "system", "content": system_message}]
            new_msg_history = other_msg + [
                {
                    "role": "assistant",
                    "content": [
                        {
                            "type": "text",
                            "text": content,
                        }
                    ],
                }
            ]    
            with open(f"{self.path2save}/experiment_log.json", 'w') as json_file:
                json.dump(system_message_json + new_msg_history, json_file, indent=4)
            print(f"Data has been saved to [{self.path2save}/experiment_log.json]")
        return content


    def call_model_for_sketch_generation(self):
        print("Calling LLM...")
        
        add_args = {}
        add_args["stop"] = "</answer>" 

        msg_history = []
        init_canvas_str = None # self.init_canvas_str

        all_llm_output = self.get_response_from_llm(
            msg=self.input_prompt,
            system_message=system_prompt.format(res=self.res),
            msg_history=msg_history,
            init_canvas_str=init_canvas_str,
            #seed_mode=self.seed_mode,
            gen_mode=self.gen_mode,
            **add_args
        )

        all_llm_output += f"</answer>"
        return all_llm_output
        

    def parse_model_to_svg(self, model_rep_sketch):
        # Parse model_rep with xml
        strokes_list_str, t_values_str = utils.parse_xml_string(model_rep_sketch, self.res)
        strokes_list, t_values = ast.literal_eval(strokes_list_str), ast.literal_eval(t_values_str)

        # extract control points from sampled lists
        all_control_points = utils.get_control_points(strokes_list, t_values, self.cells_to_pixels_map)

        # define SVG based on control point
        sketch_text_svg = utils.format_svg(all_control_points, dim=self.grid_size, stroke_width=self.stroke_width)
        return sketch_text_svg
        

    def generate_sketch(self):
        sketching_commands = self.call_model_for_sketch_generation()
        model_strokes_svg = self.parse_model_to_svg(sketching_commands)
        # saved the SVG sketch
        with open(f"{self.path2save}/{self.target_concept}.svg", "w") as svg_file:
            svg_file.write(model_strokes_svg)

        # vector->pixel 
        # save the sketch to png with blank backgournd
        cairosvg.svg2png(url=f"{self.path2save}/{self.target_concept}.svg", write_to=f"{self.path2save}/{self.target_concept}.png", background_color="white")
        
        # save the sketch to png on the canvas
        output_png_path = f"{self.path2save}/{self.target_concept}_canvas.png"
        cairosvg.svg2png(url=f"{self.path2save}/{self.target_concept}.svg", write_to=output_png_path)
        foreground = Image.open(output_png_path)
        self.init_canvas.paste(Image.open(output_png_path), (0, 0), foreground) 
        self.init_canvas.save(output_png_path)

        

# Initialize and run the SketchApp
if __name__ == '__main__':
    args = call_argparse()
    sketch_app = SketchApp(args)
    for attempts in range(3):
        try:
            sketch_app.generate_sketch()
            exit(0)
        except Exception as e:
            print(f"An error has occurred: {e}")
            traceback.print_exc()