update app.py

app.py

@@ -129,10 +129,7 @@ ram_model = None
 kosmos_model = None
 kosmos_processor = None
 
-brush_color = "#00FFFF"
-
 MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
 
 def load_model_hf(model_config_path, repo_id, filename, device='cpu'):
     args = SLConfig.fromfile(model_config_path) 
@@ -419,7 +416,7 @@ def lama_cleaner_process(image, mask, cleaner_size_limit=1080):
         
         logger.info(f'_______lama_cleaner_process_______9____')
         if config.sd_seed == -1:
-            config.sd_seed = random.randint(1, 999999999)
+            config.sd_seed = random.randint(1, MAX_SEED)
 
         # logger.info(f"Origin image shape_0_: {original_shape} / {size_limit}")
         logger.info(f'_______lama_cleaner_process_______10____')
@@ -618,29 +615,53 @@ def get_time_cost(run_task_time, time_cost_str):
 def load_kolors_inpainting(inpaint_prompt, input_image, mask_image):
     from gradio_client import Client, handle_file
     import tempfile
+    MAX_IMAGE_SIZE = 1024
     try:
+        # logger.info(f'load_kolors_inpainting_input_image={inpaint_prompt} // {input_image}')
+        # logger.info(f'load_kolors_inpainting_mask_image={mask_image}')
+
         job_image = {}
         if 'background' in input_image.keys():
+            width, height = input_image['background'].size
+            if max(width, height) > MAX_IMAGE_SIZE:
+                if width > height:
+                    resize_width = MAX_IMAGE_SIZE
+                    resize_height = int(height * MAX_IMAGE_SIZE / width)
+                else:
+                    resize_height = MAX_IMAGE_SIZE
+                    resize_width = int(width * MAX_IMAGE_SIZE / height)
+            else:
+                resize_width, resize_height = width, height
+            
+            logger.info(f"resize____{width}, {height}==>{resize_width}, {resize_height}")
             _, temp_file_path = tempfile.mkstemp(suffix='.png')
-            img = input_image['background'].convert("RGB")
+            img = input_image['background'].convert("RGB").resize((resize_width, resize_height))
             img.save(temp_file_path)
+            # logger.info(f'load_kolors_inpainting_temp_file_background_={temp_file_path}')
             job_image["background"] = handle_file(temp_file_path)
         if 'layers' in input_image.keys() and len(input_image['layers']) > 0:
             _, temp_file_path = tempfile.mkstemp(suffix='.png')
-            img = input_image['layers'][0].convert("RGB")
+            img = input_image['layers'][0].convert("RGB").resize((resize_width, resize_height))
             img.save(temp_file_path)
+            # logger.info(f'load_kolors_inpainting_temp_file_layers_={temp_file_path}')
             job_image["layers"] = [handle_file(temp_file_path)]
+
         if 'composite' in input_image.keys():
             _, temp_file_path = tempfile.mkstemp(suffix='.png')
-            img = input_image['composite'].convert("RGB")
+            img = input_image['composite'].convert("RGB").resize((resize_width, resize_height))
             img.save(temp_file_path)
+            # logger.info(f'load_kolors_inpainting_temp_file_composite_={temp_file_path}')
             job_image["composite"] = handle_file(temp_file_path)
 
         _, temp_file_path = tempfile.mkstemp(suffix='.png')
-        img = mask_image.convert("RGB")
+        img = mask_image.convert("RGB").resize((resize_width, resize_height))
         img.save(temp_file_path)
+        # logger.info(f'load_kolors_inpainting_temp_file_mask_={temp_file_path}')
         job_mask_image = handle_file(temp_file_path)
 
+        # logger.info(f'load_kolors_inpainting_job_image={job_image}')
+        # logger.info(f'load_kolors_inpainting_job_mask_image={job_mask_image}')
+
         client = Client("Kwai-Kolors/Kolors-Inpainting")
         
         job = client.submit(
@@ -648,7 +669,7 @@ def load_kolors_inpainting(inpaint_prompt, input_image, mask_image):
                 image=job_image,
                 mask_image=job_mask_image,
                 negative_prompt="broken fingers, deformed fingers, deformed hands, stumps, blurriness, low quality",
-                seed=0,
+                seed=random.randint(1, MAX_SEED), #0,
                 randomize_seed=True,
                 guidance_scale=6,
                 num_inference_steps=25,
@@ -657,7 +678,12 @@ def load_kolors_inpainting(inpaint_prompt, input_image, mask_image):
         while not job.done():
             time.sleep(0.1)
 
-        result = job.outputs()[0]
+        result = job.outputs()
+        logger.info(f'load_kolors_inpainting_result={result}')
+        if len(result) <= 0:
+            return None
+
+        result = result[0]
         im = Image.open(result)
         if im.mode == "RGBA":
             im.load()  
@@ -839,8 +865,8 @@ def run_anything_task(input_image, text_prompt, task_type, inpaint_prompt, box_t
 
         if task_type in ['inpainting', 'outpainting']:
             # inpainting pipeline
-            image_source_for_inpaint = image_pil.resize((512, 512))
-            image_mask_for_inpaint = mask_pil.resize((512, 512))
+            image_source_for_inpaint = image_pil #.resize((512, 512))
+            image_mask_for_inpaint = mask_pil #.resize((512, 512))
             if task_type in ['outpainting']:
                 # reverse mask
                 img_arr = np.array(image_mask_for_inpaint)
@@ -973,23 +999,26 @@ def main_gradio(args):
                     task_types.append("segment")
                 if inpainting_enable:
                     task_types.append("inpainting")
-                    task_types.append("outpainting")
+                    # task_types.append("outpainting")
                 if lama_cleaner_enable:
                     task_types.append("remove")
                 if ram_enable:
                     task_types.append("relate anything")
                 if kosmos_enable:
                     task_types.append("Kosmos-2")           
-
-                input_image = gr.ImageMask(sources='upload', elem_id="image_upload", type='pil', label="Upload",
-                                    brush=gr.Brush(colors=[brush_color], color_mode="fixed"))
+                
+                brush_color = "#00FF00"
+                color_mode = "fixed"
+                input_image = gr.ImageMask(sources=["upload", "webcam"], image_mode='RGB', elem_id="image_upload", type='pil', label="Upload",
+                                   brush=gr.Brush(colors=[brush_color], color_mode=color_mode))
+                
                 task_type = gr.Radio(task_types,  value="detection", 
                                                 label='Task type', visible=True) 
                 mask_source_radio = gr.Radio([mask_source_draw, mask_source_segment], 
                                     value=mask_source_segment, label="Mask from",
                                     visible=False) 
                 text_prompt = gr.Textbox(label="Detection Prompt[To detect multiple objects, seperating each with '.', like this: cat . dog . chair ]", placeholder="Cannot be empty")                                                
-                inpaint_prompt = gr.Textbox(label="Inpaint/Outpaint Prompt (if this is empty, then remove)", visible=False)
+                inpaint_prompt = gr.Textbox(label="Inpaint Prompt (if this is empty, then remove)", visible=False)
                 num_relation = gr.Slider(label="How many relations do you want to see", minimum=1, maximum=20, value=5, step=1, visible=False)
                 
                 kosmos_input = gr.Radio(["Brief", "Detailed"], label="Kosmos Description Type", value="Brief", visible=False)