added new features

app.py

@@ -7,6 +7,7 @@ import os.path as osp
 import random
 import numpy.random as npr
 import sys
+import imageio
 
 # sys.path.append('./code')
 
@@ -29,7 +30,7 @@ from diffusers import StableDiffusionPipeline
 
 device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
 
-# model = None 
+model = None 
 model = StableDiffusionPipeline.from_pretrained("runwayml/stable-diffusion-v1-5").to(device)
 
 from typing import Mapping
@@ -56,7 +57,6 @@ TITLE="""<h1 style="font-size: 42px;" align="center">Word-To-Image: Morphing Ara
 DESCRIPTION="""This demo builds on the [Word-As-Image for Semantic Typography](https://wordasimage.github.io/Word-As-Image-Page/) work to support Arabic fonts and morphing whole words into semantic concepts. It is part of an ongoing project with the [ARBML](https://arbml.github.io/website/) community."""
 
 # DESCRIPTION += '\n<p>This demo is licensed under a <a rel="license" href="http://creativecommons.org/licenses/by-sa/4.0/"> Creative Commons Attribution-ShareAlike 4.0 International License</a>.</p>'
-# DESCRIPTION += """<br>For faster inference without waiting in queue, you can [![]()]()"""
 DESCRIPTION += '\n<p>For faster inference without waiting in queue, you can <a href="https://colab.research.google.com/drive/1wobOAsnLpkIzaRxG5yac8NcV7iCrlycP"><img style="display: inline; margin-top: 0em; margin-bottom: 0em" src="https://colab.research.google.com/assets/colab-badge.svg" alt="Open In Colab"/></a></p>'
 
 if (SPACE_ID := os.getenv('SPACE_ID')) is not None:
@@ -74,7 +74,7 @@ pydiffvg.set_print_timing(False)
 gamma = 1.0
 
 
-def set_config(semantic_concept, word, prompt, font_name, num_steps):
+def set_config(semantic_concept, word, prompt_suffix, font_name, num_steps, seed, dist_loss_weight, pixel_dist_kernel_blur, pixel_dist_sigma, angeles_w):
     
     cfg_d = edict()
     cfg_d.config = "code/config/base.yaml"
@@ -95,16 +95,22 @@ def set_config(semantic_concept, word, prompt, font_name, num_steps):
     del cfgs
 
     cfg.semantic_concept = semantic_concept
+    cfg.prompt_suffix = prompt_suffix
     cfg.word = word
     cfg.optimized_letter = word
     cfg.font = font_name
-    cfg.seed = 0
+    cfg.seed = seed
     cfg.num_iter = num_steps
     cfg.batch_size = 1
+    cfg.loss.tone.dist_loss_weight = dist_loss_weight
+    cfg.loss.tone.pixel_dist_kernel_blur = pixel_dist_kernel_blur
+    cfg.loss.tone.pixel_dist_sigma = pixel_dist_sigma
+    cfg.loss.conformal.angeles_w = angeles_w
     
-    if ' ' in cfg.word:
-        raise gr.Error(f'should be only one word')
-    cfg.caption = prompt
+    # if ' ' in cfg.word:
+    #     raise gr.Error(f'should be only one word')
+
+    cfg.caption = f"a {cfg.semantic_concept}. {cfg.prompt_suffix}"
     cfg.log_dir = f"output/{cfg.experiment}_{cfg.word}"
     if cfg.optimized_letter in cfg.word:
         cfg.optimized_letter = cfg.optimized_letter
@@ -151,14 +157,14 @@ def init_shapes(svg_path, trainable: Mapping[str, bool]):
     return shapes_init, shape_groups_init, parameters
 
 
-def run_main_ex(word, semantic_concept, num_steps):
-    prompt = f"a {semantic_concept}. minimal flat 2d vector. lineal color. trending on artstation"
+def run_main_ex(word, semantic_concept, num_steps, seed):
+    prompt_suffix = "minimal flat 2d vector. lineal color. trending on artstation"
     font_name = "ArefRuqaa"
-    return list(next(run_main_app(semantic_concept, word, prompt, font_name, num_steps, 0)))
+    return list(next(run_main_app(semantic_concept, word, prompt_suffix, font_name, num_steps, seed, 100, 201, 30, 0.5, 0)))
                         
-def run_main_app(semantic_concept, word, prompt, font_name, num_steps, example=0):
+def run_main_app(semantic_concept, word, prompt_suffix, font_name, num_steps, seed, dist_loss_weight, pixel_dist_kernel_blur, pixel_dist_sigma, angeles_w, example=0):
     
-    cfg = set_config(semantic_concept, word, prompt, font_name, num_steps)
+    cfg = set_config(semantic_concept, word, prompt_suffix, font_name, num_steps, seed, dist_loss_weight, pixel_dist_kernel_blur, pixel_dist_sigma, angeles_w)
 
     pydiffvg.set_use_gpu(torch.cuda.is_available())
 
@@ -204,6 +210,7 @@ def run_main_app(semantic_concept, word, prompt, font_name, num_steps, example=0
     print("start training")
     # training loop
     t_range = tqdm(range(num_iter))
+    gif_frames = []
     for step in t_range:
         optim.zero_grad()
 
@@ -215,9 +222,10 @@ def run_main_app(semantic_concept, word, prompt, font_name, num_steps, example=0
         img = img[:, :, 3:4] * img[:, :, :3] + torch.ones(img.shape[0], img.shape[1], 3, device=device) * (
                     1 - img[:, :, 3:4])
         img = img[:, :, :3]
+        gif_frames += [img]
 
-        filename = os.path.join(
-            cfg.experiment_dir, "video-svg", f"iter{step:04d}.svg")
+        
+        filename = os.path.join(cfg.experiment_dir, "video-svg", f"iter{step:04d}.svg")
         check_and_create_dir(filename)
         save_svg.save_svg(filename, w, h, shapes, shape_groups)
         if not example:
@@ -250,8 +258,10 @@ def run_main_app(semantic_concept, word, prompt, font_name, num_steps, example=0
 
     combine_word(cfg.word, cfg.optimized_letter, cfg.font, cfg.experiment_dir, device)
 
-    image = os.path.join(cfg.experiment_dir,f"{cfg.font}_{cfg.word}_{cfg.optimized_letter}.svg")
-    yield gr.update(value=filename_init,visible=True),gr.update(visible=True),gr.update(value=image,visible=True)
+    filename = os.path.join(cfg.experiment_dir, "final.gif")
+    imageio.mimsave(filename, gif_frames)
+
+    yield gr.update(value=filename_init,visible=True),gr.update(visible=False),gr.update(value=filename,visible=True)
  
 
 def change_prompt(concept, prompt_suffix):
@@ -294,6 +304,37 @@ with gr.Blocks() as demo:
                 value="a {concept}. minimal flat 2d vector. lineal color. trending on artstation."   
             )
 
+            with gr.Row():
+
+                with gr.Accordion("Advanced Parameters", open=False, visible=True):
+    
+                    seed = gr.Number(
+                        label='Seed',
+                        value=42   
+                    )
+
+                    angeles_w = gr.Number(
+                        label='ACAP Deformation Loss Weight',
+                        value=0.5   
+                    )
+
+                    dist_loss_weight = gr.Number(
+                        label='Tone Loss: dist_loss_weight',
+                        value=100   
+                    )
+
+                    pixel_dist_kernel_blur = gr.Number(
+                        label='Tone Loss: pixel_dist_kernel_blur',
+                        value=201   
+                    )
+
+                    pixel_dist_sigma = gr.Number(
+                        label='Tone Loss: pixel_dist_sigma',
+                        value=30   
+                    )
+
+
+
             semantic_concept.change(change_prompt, [semantic_concept, prompt_suffix], prompt)
             prompt_suffix.change(change_prompt, [semantic_concept, prompt_suffix], prompt)
 
@@ -301,7 +342,7 @@ with gr.Blocks() as demo:
                       minimum=0,
                       maximum=500,
                       step=10,
-                      value=500)
+                      value=250)
             
             font_name = gr.Text(value=None,visible=False,label="Font Name")
         
@@ -314,25 +355,26 @@ with gr.Blocks() as demo:
             run = gr.Button('Generate')
 
         with gr.Column():
-            result0 = gr.Image(type="filepath", label="Initial Word").style(height=170)
+            result0 = gr.Image(type="filepath", label="Initial Word").style(height=250)
             result1 = gr.Image(type="filepath", label="Optimization Process").style(height=300)
-            result2 = gr.Image(type="filepath", label="Final Result",visible=False).style(height=100)
+            result2 = gr.Image(type="filepath", label="Final Result",visible=False).style(height=300)
             
         
     with gr.Row():
         # examples
         examples = [
-            ["قطة", "Cat", 500],
-            ["كلب", "Dog", 500],
-            ["حصان", "Horse", 500],
-            ["أخطبوط", "Octopus", 500],
+            ["قطة", "Cat", 250, 42],
+            ["كلب", "Dog", 250, 42],
+            ["حصان", "Horse", 250, 42],
+            ["أخطبوط", "Octopus", 250, 42],
         ]
-        demo.queue(max_size=10, concurrency_count=2)
+        demo.queue(max_size=10, concurrency_count=1)
         gr.Examples(examples=examples,
                 inputs=[
                     word,
                     semantic_concept,
-                    num_steps
+                    num_steps,
+                    seed
                 ],
                 outputs=[
                     result0,
@@ -347,9 +389,14 @@ with gr.Blocks() as demo:
     inputs = [
         semantic_concept,
         word, 
-        prompt,
+        prompt_suffix,
         font_name,
-        num_steps
+        num_steps,
+        seed,
+        dist_loss_weight,
+        pixel_dist_kernel_blur,
+        pixel_dist_sigma,
+        angeles_w
     ]
 
     outputs = [

code/config.py

@@ -40,8 +40,8 @@ def parse_args():
     cfg.font = args.font
     cfg.semantic_concept = args.semantic_concept
     cfg.word = cfg.semantic_concept if args.word == "none" else args.word
-    if " " in cfg.word:
-      raise ValueError(f'no spaces are allowed')
+    # if " " in cfg.word:
+    #   raise ValueError(f'no spaces are allowed')
     if "jpeg" in args.semantic_concept:
         cfg.caption = args.semantic_concept 
     else:

requirements.txt

@@ -5,6 +5,7 @@ torchvision==0.13.1+cu113
 cmake
 numpy
 scikit-image
+imageio
 ffmpeg
 svgwrite
 svgpathtools