Update app.py

app.py

@@ -1,142 +1,114 @@
 import gradio as gr
 import numpy as np
-import random
-#import spaces #[uncomment to use ZeroGPU]
-from diffusers import DiffusionPipeline
+#import random
+import spaces #[uncomment to use ZeroGPU]
+#from diffusers import DiffusionPipeline
 import torch
 
-device = "cuda" if torch.cuda.is_available() else "cpu"
-model_repo_id = "stabilityai/sdxl-turbo" #Replace to the model you would like to use
+from diffusers import AutoPipelineForInpainting
+from diffusers.utils import load_image
+from transformers import AutoModelForMaskGeneration, AutoProcessor, pipeline
 
-if torch.cuda.is_available():
-    torch_dtype = torch.float16
-else:
-    torch_dtype = torch.float32
+#import cv2
+#import matplotlib.pyplot as plt
+from PIL import Image
+import os
+import gc
 
-pipe = DiffusionPipeline.from_pretrained(model_repo_id, torch_dtype=torch_dtype)
-pipe = pipe.to(device)
 
-MAX_SEED = np.iinfo(np.int32).max
-MAX_IMAGE_SIZE = 1024
+DEVICE = "cuda" if torch.cuda.is_available() else "cpu"
 
-#@spaces.GPU #[uncomment to use ZeroGPU]
-def infer(prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps, progress=gr.Progress(track_tqdm=True)):
+GDINO_MODEL_NAME="IDEA-Research/grounding-dino-tiny" 
+SAM_MODEL_NAME="facebook/sam-vit-base" 
 
-    if randomize_seed:
-        seed = random.randint(0, MAX_SEED)
-        
-    generator = torch.Generator().manual_seed(seed)
+GDINO=pipeline(model=GDINO_MODEL_NAME, task="zero-shot-object-detection", device=DEVICE)
+SAM=AutoModelForMaskGeneration.from_pretrained(SAM_MODEL_NAME).to(DEVICE)
+SAM_PROCESSOR=AutoProcessor.from_pretrained(SAM_MODEL_NAME)
+
+SD_MODEL="diffusers/stable-diffusion-xl-1.0-inpainting-0.1" 
+SD_PIPLINE = AutoPipelineForInpainting.from_pretrained(SD_MODEL, torch_dtype=torch.float16).to(DEVICE)
+IP_ADAPTER="h94/IP-Adapter" 
+SUB_FOLDER="sdxl_models"  
+IP_WEIGHT_NAME="ip-adapter_sdxl.bin" 
+SD_PIPLINE.load_ip_adapter(IP_ADAPTER, subfolder=SUB_FOLDER, weight_name=IP_WEIGHT_NAME)
+IP_SCALE=0.6 
+SD_PIPLINE.set_ip_adapter_scale(IP_SCALE)
+
+GEN_STEPS=100 
+
+
+def refine_masks(masks: torch.BoolTensor)->np.array:    
+    masks = masks.permute(0, 2, 3, 1)
+    masks = masks.float().mean(axis=-1)
+    return masks.cpu().numpy()  
+
+
+def get_boxes(detections:list)-> list:
+    boxes = []
+    for det in detections:
+        boxes.append([det['box']['xmin'], det['box']['ymin'], 
+                      det['box']['xmax'], det['box']['ymax']])
+    return [boxes]
+
+
+def get_mask(img:Image, prompt:str, d_model:pipeline, s_model:AutoModelForMaskGeneration,
+             s_processor:AutoProcessor, device:str, threshold:float = 0.3)-> np.array:
     
-    image = pipe(
-        prompt = prompt, 
-        negative_prompt = negative_prompt,
-        guidance_scale = guidance_scale, 
-        num_inference_steps = num_inference_steps, 
-        width = width, 
-        height = height,
-        generator = generator
-    ).images[0] 
+    labels = [label if label.endswith(".") else label+"." for label in ['face', prompt]] 
+    dets=d_model(img, candidate_labels=labels, threshold=threshold)    
+
+    boxes = get_boxes(dets)
+    inputs=s_processor(images=img, input_boxes=boxes, return_tensors="pt").to(DEVICE)
+    outputs = s_model(**inputs)
     
-    return image, seed
-
-examples = [
-    "Astronaut in a jungle, cold color palette, muted colors, detailed, 8k",
-    "An astronaut riding a green horse",
-    "A delicious ceviche cheesecake slice",
-]
-
-css="""
-#col-container {
-    margin: 0 auto;
-    max-width: 640px;
-}
-"""
-
-with gr.Blocks(css=css) as demo:
+    masks = s_processor.post_process_masks(
+        masks=outputs.pred_masks,
+        original_sizes=inputs.original_sizes,
+        reshaped_input_sizes=inputs.reshaped_input_sizes
+    )[0] 
+
+    return refine_masks(masks) 
+
+
+def generate_result(model_img:str, cloth_img:str,
+                    masks: np.array, prompt:str, sd_pipline:AutoPipelineForInpainting, n_steps:int=100)->Image:
     
-    with gr.Column(elem_id="col-container"):
-        gr.Markdown(f"""
-        # Text-to-Image Gradio Template
-        """)
-        
-        with gr.Row():
-            
-            prompt = gr.Text(
-                label="Prompt",
-                show_label=False,
-                max_lines=1,
-                placeholder="Enter your prompt",
-                container=False,
-            )
-            
-            run_button = gr.Button("Run", scale=0)
-        
-        result = gr.Image(label="Result", show_label=False)
-
-        with gr.Accordion("Advanced Settings", open=False):
-            
-            negative_prompt = gr.Text(
-                label="Negative prompt",
-                max_lines=1,
-                placeholder="Enter a negative prompt",
-                visible=False,
-            )
-            
-            seed = gr.Slider(
-                label="Seed",
-                minimum=0,
-                maximum=MAX_SEED,
-                step=1,
-                value=0,
-            )
-            
-            randomize_seed = gr.Checkbox(label="Randomize seed", value=True)
-            
-            with gr.Row():
-                
-                width = gr.Slider(
-                    label="Width",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024, #Replace with defaults that work for your model
-                )
-                
-                height = gr.Slider(
-                    label="Height",
-                    minimum=256,
-                    maximum=MAX_IMAGE_SIZE,
-                    step=32,
-                    value=1024, #Replace with defaults that work for your model
-                )
-            
-            with gr.Row():
-                
-                guidance_scale = gr.Slider(
-                    label="Guidance scale",
-                    minimum=0.0,
-                    maximum=10.0,
-                    step=0.1,
-                    value=0.0, #Replace with defaults that work for your model
-                )
-                
-                num_inference_steps = gr.Slider(
-                    label="Number of inference steps",
-                    minimum=1,
-                    maximum=50,
-                    step=1,
-                    value=2, #Replace with defaults that work for your model
-                )
+    width, height = model_img.size
+    
+    cloth_mask=masks[1] #np.array(masks[1],dtype=np.float32)
+    generator = torch.Generator(device="cpu").manual_seed(4)
+    images = sd_pipline(
+        prompt=prompt,
+        image=model_img, 
+        mask_image=cloth_mask,
+        ip_adapter_image=cloth_img, 
+        generator=generator,
+        num_inference_steps=n_steps,
         
-        gr.Examples(
-            examples = examples,
-            inputs = [prompt]
-        )
-    gr.on(
-        triggers=[run_button.click, prompt.submit],
-        fn = infer,
-        inputs = [prompt, negative_prompt, seed, randomize_seed, width, height, guidance_scale, num_inference_steps],
-        outputs = [result, seed]
-    )
-
-demo.queue().launch()
+    ).images
+    
+    return images[0].resize((width, height))
+
+
+@spaces.GPU
+def run(model_img:Image, cloth_img:Image, cloth_class:str, close_description:str)->Image:
+    masks = get_mask(model_img, cloth_class, GDINO, SAM, SAM_PROCESSOR, DEVICE) #GSAM2)
+    result = generate_result(model_img, cloth_img, masks, close_description, SD_PIPLINE, GEN_STEPS)
+    gc.collect()
+    torch.cuda.empty_cache()
+    return result
+
+
+gr.Interface(
+    run,
+    title = 'Virtual Try-On',
+    inputs=[
+        gr.Image(sources = 'upload', label='Model image', type = 'pil'),
+        gr.Image(sources = 'upload', label='Cloth image', type = 'pil'),
+        gr.Textbox(label = 'Cloth class'),
+        gr.Textbox(label = 'Close description')
+    ],
+    outputs = [
+        gr.Image()
+        ]
+).launch(debug=True,share=True)