[README] Update

README.md

@@ -80,12 +80,38 @@ pip install -r requirements.txt
 ```
 
 ## 🏋️ Training
+### Training - Phase 1
+```bash
+sh train_phase_1.sh
+```
+
+### Training - Phase 2
+Coming Soon...
 
 ## 📺 Sampling
-coming soon ...
+### Step 1 => Prepare the checkpoint   
+(1) Download the checkpoint or (2) Put your checkpoint to the folder `ckpt/`
+
+### Step 2 => Run the script  
+(1) Sampling image from content image.  
+```bash
+sh script/sample_content_image.sh
+```
+(2) Sampling image from content character.
+```bash
+sh script/sample_content_character.sh
+```
 
 ## 📱 Run WebUI
-coming soon ...
+### (1) Sampling by FontDiffuser
+```bash
+gradio gradio_app.py
+```
+
+### (2) Sampling by FontDiffuser and Rendering by ControlNet
+```bash
+gradio gradio_app_controlnet.py
+```
 
 ## 🌄 Gallery
 coming sonn ...

sample.py

@@ -58,8 +58,10 @@ def image_process(args, content_image=None, style_image=None):
                 return None, None
             font = load_ttf(ttf_path=args.ttf_path)
             content_image = ttf2im(font=font, char=args.content_character)
+            content_image_pil = content_image.copy()
         else:
             content_image = Image.open(args.content_image_path).convert('RGB')
+            content_image_pil = None
         style_image = Image.open(args.style_image_path).convert('RGB')
     else:
         assert style_image is not None, "The style image should not be None."
@@ -86,7 +88,7 @@ def image_process(args, content_image=None, style_image=None):
     content_image = content_inference_transforms(content_image)[None, :]
     style_image = style_inference_transforms(style_image)[None, :]
 
-    return content_image, style_image
+    return content_image, style_image, content_image_pil
 
 def load_fontdiffuer_pipeline(args):
     # Load the model state_dict
@@ -129,9 +131,9 @@ def sampling(args, pipe, content_image=None, style_image=None):
     if args.seed:
         set_seed(seed=args.seed)
 
-    content_image, style_image = image_process(args=args,
-                                               content_image=content_image,
-                                               style_image=style_image)
+    content_image, style_image, content_image_pil = image_process(args=args, 
+                                                                  content_image=content_image, 
+                                                                  style_image=style_image)
     if content_image == None:
         print(f"The content_character you provided is not in the ttf. \
                 Please change the content_character or you can change the ttf.")
@@ -161,11 +163,20 @@ def sampling(args, pipe, content_image=None, style_image=None):
         if args.save_image:
             print(f"Saving the image ......")
             save_single_image(save_dir=args.save_image_dir, image=images[0])
-            save_image_with_content_style(save_dir=args.save_image_dir,
-                                        image=images[0],
-                                        content_image_path=args.content_image_path,
-                                        style_image_path=args.style_image_path,
-                                        resolution=args.resolution)
+            if args.character_input:
+                save_image_with_content_style(save_dir=args.save_image_dir,
+                                            image=images[0],
+                                            content_image_pil=content_image_pil,
+                                            content_image_path=None,
+                                            style_image_path=args.style_image_path,
+                                            resolution=args.resolution)
+            else:
+                save_image_with_content_style(save_dir=args.save_image_dir,
+                                            image=images[0],
+                                            content_image_pil=None,
+                                            content_image_path=args.content_image_path,
+                                            style_image_path=args.style_image_path,
+                                            resolution=args.resolution)
             print(f"Finish the sampling process, costing time {end - start}s")
         return images[0]
 

scripts/sample_content_character.sh

@@ -0,0 +1,13 @@
+python sample.py \
+    --ckpt_dir="ckpt/" \
+    --style_image_path="data_examples/sampling/example_style.jpg" \
+    --save_image \
+    --character_input \
+    --content_character="隆" \
+    --save_image_dir="outputs/" \
+    --device="cuda:0" \
+    --algorithm_type="dpmsolver++" \
+    --guidance_type="classifier-free" \
+    --guidance_scale=7.5 \
+    --num_inference_steps=20 \
+    --method="multistep"

scripts/sample.sh → scripts/sample_content_image.sh

@@ -2,6 +2,7 @@ python sample.py \
     --ckpt_dir="ckpt/" \
     --content_image_path="data_examples/sampling/example_content.jpg" \
     --style_image_path="data_examples/sampling/example_style.jpg" \
+    --save_image \
     --save_image_dir="outputs/" \
     --device="cuda:0" \
     --algorithm_type="dpmsolver++" \

utils.py

@@ -25,10 +25,13 @@ def save_single_image(save_dir, image):
     image.save(save_path)
 
 
-def save_image_with_content_style(save_dir, image, content_image_path, style_image_path, resolution):
+def save_image_with_content_style(save_dir, image, content_image_pil, content_image_path, style_image_path, resolution):
 
     new_image = Image.new('RGB', (resolution*3, resolution))
-    content_image = Image.open(content_image_path).convert("RGB").resize((resolution, resolution), Image.BILINEAR)
+    if content_image_pil is not None:
+        content_image = content_image_pil
+    else:
+        content_image = Image.open(content_image_path).convert("RGB").resize((resolution, resolution), Image.BILINEAR)
     style_image = Image.open(style_image_path).convert("RGB").resize((resolution, resolution), Image.BILINEAR)
 
     new_image.paste(content_image, (0, 0))