Add: diffsketcher handler.py with original implementation
Browse files- handler.py +44 -85
handler.py
CHANGED
@@ -36,83 +36,58 @@ try:
|
|
36 |
except ImportError as e:
|
37 |
debug_log(f"Error importing DiffSketcher models: {e}")
|
38 |
debug_log(traceback.format_exc())
|
|
|
39 |
|
40 |
class EndpointHandler:
|
41 |
def __init__(self, model_dir):
|
42 |
"""Initialize the handler with model directory"""
|
43 |
-
|
44 |
-
|
45 |
-
|
46 |
-
|
47 |
-
|
48 |
-
|
49 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
50 |
try:
|
51 |
-
|
52 |
-
|
53 |
-
|
54 |
-
|
55 |
-
|
56 |
-
|
57 |
-
|
58 |
-
|
59 |
-
|
60 |
-
|
61 |
-
|
62 |
-
self.use_model = True
|
63 |
-
else:
|
64 |
-
debug_log(f"Checkpoint not found at {weights_path}, using model without pre-trained weights")
|
65 |
-
self.use_model = True
|
66 |
except Exception as e:
|
67 |
-
debug_log(f"Error
|
68 |
debug_log(traceback.format_exc())
|
69 |
-
|
70 |
-
except Exception as e:
|
71 |
-
debug_log(f"Error in handler initialization: {e}")
|
72 |
-
debug_log(traceback.format_exc())
|
73 |
-
self.use_model = False
|
74 |
|
75 |
def generate_svg(self, prompt, width=512, height=512):
|
76 |
"""Generate an SVG from a text prompt"""
|
77 |
debug_log(f"Generating SVG for prompt: {prompt}")
|
78 |
|
79 |
-
|
80 |
-
|
81 |
-
|
82 |
-
|
83 |
-
|
84 |
-
|
85 |
-
latent = self.diffusion_model.generate(text_features)
|
86 |
-
svg_data = self.sketch_model.generate(latent, num_paths=20, width=width, height=height)
|
87 |
-
debug_log("Generated SVG using DiffSketcher")
|
88 |
-
return svg_data
|
89 |
-
except Exception as e:
|
90 |
-
debug_log(f"Error generating SVG with model: {e}")
|
91 |
-
debug_log(traceback.format_exc())
|
92 |
-
return self._generate_placeholder_svg(prompt, width, height)
|
93 |
-
else:
|
94 |
-
debug_log("Model not initialized, using placeholder")
|
95 |
-
return self._generate_placeholder_svg(prompt, width, height)
|
96 |
-
|
97 |
-
def _generate_placeholder_svg(self, prompt, width=512, height=512):
|
98 |
-
"""Generate a placeholder SVG"""
|
99 |
-
debug_log(f"Generating placeholder SVG for prompt: {prompt}")
|
100 |
-
|
101 |
-
# Create a more interesting placeholder that looks like a sketch
|
102 |
-
svg_content = f"""<svg width="{width}" height="{height}" xmlns="http://www.w3.org/2000/svg">
|
103 |
-
<rect width="100%" height="100%" fill="#ffffff"/>
|
104 |
-
<g stroke="#000000" fill="none">
|
105 |
-
<!-- Draw a simple sketch based on the prompt -->
|
106 |
-
<circle cx="{width/2}" cy="{height/2}" r="{min(width, height)/4}" stroke-width="2"/>
|
107 |
-
<ellipse cx="{width/2}" cy="{height/2}" rx="{width/3}" ry="{height/4}" stroke-width="1.5"/>
|
108 |
-
<path d="M {width/4} {height/4} Q {width/2} {height/8} {3*width/4} {height/4}" stroke-width="2"/>
|
109 |
-
<path d="M {width/4} {3*height/4} Q {width/2} {7*height/8} {3*width/4} {3*height/4}" stroke-width="2"/>
|
110 |
-
</g>
|
111 |
-
<text x="50%" y="50%" font-family="Arial" font-size="20" text-anchor="middle" fill="#333333">{prompt}</text>
|
112 |
-
</svg>"""
|
113 |
-
|
114 |
-
debug_log("Generated placeholder SVG")
|
115 |
-
return svg_content
|
116 |
|
117 |
def __call__(self, data):
|
118 |
"""Handle a request to the model"""
|
@@ -136,19 +111,9 @@ class EndpointHandler:
|
|
136 |
svg_content = self.generate_svg(prompt)
|
137 |
|
138 |
# Convert SVG to PNG
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
debug_log("Generated image from SVG")
|
143 |
-
except Exception as e:
|
144 |
-
debug_log(f"Error converting SVG to PNG: {e}")
|
145 |
-
debug_log(traceback.format_exc())
|
146 |
-
# Create a simple placeholder image
|
147 |
-
image = Image.new("RGB", (512, 512), color="#f0f0f0")
|
148 |
-
from PIL import ImageDraw
|
149 |
-
draw = ImageDraw.Draw(image)
|
150 |
-
draw.text((256, 256), prompt, fill="black", anchor="mm")
|
151 |
-
debug_log("Created placeholder image")
|
152 |
|
153 |
# Return the PIL Image directly
|
154 |
debug_log("Returning image")
|
@@ -156,10 +121,4 @@ class EndpointHandler:
|
|
156 |
except Exception as e:
|
157 |
debug_log(f"Error in handler: {e}")
|
158 |
debug_log(traceback.format_exc())
|
159 |
-
|
160 |
-
image = Image.new("RGB", (512, 512), color="#ff0000")
|
161 |
-
from PIL import ImageDraw
|
162 |
-
draw = ImageDraw.Draw(image)
|
163 |
-
draw.text((256, 256), f"Error: {str(e)}", fill="white", anchor="mm")
|
164 |
-
debug_log("Returning error image")
|
165 |
-
return image
|
|
|
36 |
except ImportError as e:
|
37 |
debug_log(f"Error importing DiffSketcher models: {e}")
|
38 |
debug_log(traceback.format_exc())
|
39 |
+
raise ImportError(f"Failed to import DiffSketcher models: {e}")
|
40 |
|
41 |
class EndpointHandler:
|
42 |
def __init__(self, model_dir):
|
43 |
"""Initialize the handler with model directory"""
|
44 |
+
debug_log(f"Initializing handler with model_dir: {model_dir}")
|
45 |
+
self.model_dir = model_dir
|
46 |
+
self.device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
|
47 |
+
debug_log(f"Using device: {self.device}")
|
48 |
+
|
49 |
+
# Initialize the model
|
50 |
+
self.clip_model = ClipModel(device=self.device)
|
51 |
+
self.diffusion_model = DiffusionModel(device=self.device)
|
52 |
+
self.sketch_model = SketchModel(device=self.device)
|
53 |
+
|
54 |
+
# Load checkpoint if available
|
55 |
+
weights_path = os.path.join(model_dir, "checkpoint.pth")
|
56 |
+
if os.path.exists(weights_path):
|
57 |
+
debug_log(f"Loading checkpoint from {weights_path}")
|
58 |
+
checkpoint = torch.load(weights_path, map_location=self.device)
|
59 |
+
self.sketch_model.load_state_dict(checkpoint['sketch_model'])
|
60 |
+
debug_log("Successfully loaded checkpoint")
|
61 |
+
else:
|
62 |
+
debug_log(f"Checkpoint not found at {weights_path}, using model without pre-trained weights")
|
63 |
+
# Download the checkpoint if not available
|
64 |
try:
|
65 |
+
debug_log("Attempting to download checkpoint...")
|
66 |
+
import urllib.request
|
67 |
+
os.makedirs(os.path.dirname(weights_path), exist_ok=True)
|
68 |
+
urllib.request.urlretrieve(
|
69 |
+
"https://github.com/ximinng/DiffSketcher/releases/download/v0.1-weights/diffvg_checkpoint.pth",
|
70 |
+
weights_path
|
71 |
+
)
|
72 |
+
debug_log(f"Downloaded checkpoint to {weights_path}")
|
73 |
+
checkpoint = torch.load(weights_path, map_location=self.device)
|
74 |
+
self.sketch_model.load_state_dict(checkpoint['sketch_model'])
|
75 |
+
debug_log("Successfully loaded downloaded checkpoint")
|
|
|
|
|
|
|
|
|
76 |
except Exception as e:
|
77 |
+
debug_log(f"Error downloading checkpoint: {e}")
|
78 |
debug_log(traceback.format_exc())
|
79 |
+
debug_log("Continuing with uninitialized weights")
|
|
|
|
|
|
|
|
|
80 |
|
81 |
def generate_svg(self, prompt, width=512, height=512):
|
82 |
"""Generate an SVG from a text prompt"""
|
83 |
debug_log(f"Generating SVG for prompt: {prompt}")
|
84 |
|
85 |
+
# Generate SVG using DiffSketcher
|
86 |
+
text_features = self.clip_model.encode_text(prompt)
|
87 |
+
latent = self.diffusion_model.generate(text_features)
|
88 |
+
svg_data = self.sketch_model.generate(latent, num_paths=20, width=width, height=height)
|
89 |
+
debug_log("Generated SVG using DiffSketcher")
|
90 |
+
return svg_data
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
91 |
|
92 |
def __call__(self, data):
|
93 |
"""Handle a request to the model"""
|
|
|
111 |
svg_content = self.generate_svg(prompt)
|
112 |
|
113 |
# Convert SVG to PNG
|
114 |
+
png_data = cairosvg.svg2png(bytestring=svg_content.encode("utf-8"))
|
115 |
+
image = Image.open(io.BytesIO(png_data))
|
116 |
+
debug_log("Generated image from SVG")
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
117 |
|
118 |
# Return the PIL Image directly
|
119 |
debug_log("Returning image")
|
|
|
121 |
except Exception as e:
|
122 |
debug_log(f"Error in handler: {e}")
|
123 |
debug_log(traceback.format_exc())
|
124 |
+
raise Exception(f"Error generating image: {str(e)}")
|
|
|
|
|
|
|
|
|
|
|
|