init

.gitignore

@@ -0,0 +1,138 @@
+# Byte-compiled / optimized / DLL files
+__pycache__/
+*.py[cod]
+*$py.class
+
+# C extensions
+*.so
+
+# Distribution / packaging
+.Python
+build/
+develop-eggs/
+dist/
+downloads/
+eggs/
+.eggs/
+lib/
+lib64/
+parts/
+sdist/
+var/
+wheels/
+pip-wheel-metadata/
+share/python-wheels/
+*.egg-info/
+.installed.cfg
+*.egg
+MANIFEST
+.docker/
+
+# PyInstaller
+#  Usually these files are written by a python script from a template
+#  before PyInstaller builds the exe, so as to inject date/other infos into it.
+*.manifest
+*.spec
+
+# Installer logs
+pip-log.txt
+pip-delete-this-directory.txt
+
+# Unit test / coverage reports
+htmlcov/
+.tox/
+.nox/
+.coverage
+.coverage.*
+.cache
+nosetests.xml
+coverage.xml
+*.cover
+*.py,cover
+.hypothesis/
+.pytest_cache/
+
+# Translations
+*.mo
+*.pot
+
+# Django stuff:
+*.log
+local_settings.py
+db.sqlite3
+db.sqlite3-journal
+
+# Flask stuff:
+instance/
+.webassets-cache
+
+# Scrapy stuff:
+.scrapy
+
+# Sphinx documentation
+docs/_build/
+
+# PyBuilder
+target/
+
+# Jupyter Notebook
+.ipynb_checkpoints
+
+# IPython
+profile_default/
+ipython_config.py
+
+# pipenv
+#   According to pypa/pipenv#598, it is recommended to include Pipfile.lock in version control.
+#   However, in case of collaboration, if having platform-specific dependencies or dependencies
+#   having no cross-platform support, pipenv may install dependencies that don't work, or not
+#   install all needed dependencies.
+#Pipfile.lock
+
+# PEP 582; used by e.g. github.com/David-OConnor/pyflow
+__pypackages__/
+
+# Celery stuff
+celerybeat-schedule
+celerybeat.pid
+
+# SageMath parsed files
+*.sage.py
+
+# Environments
+.env
+# direnv
+.envrc 
+.venv
+env/
+venv/
+ENV/
+env.bak/
+venv.bak/
+docker-compose-interpreter-local.yml
+
+# Spyder project settings
+.spyderproject
+.spyproject
+
+# Rope project settings
+.ropeproject
+
+# mkdocs documentation
+/site
+
+# mypy
+.mypy_cache/
+.dmypy.json
+dmypy.json
+
+# Pyre type checker
+.pyre/
+
+# ide
+.idea/
+.vscode/
+
+# macos
+.DS_Store
+.envrc

app.py

@@ -0,0 +1,45 @@
+import gradio as gr
+import os
+
+from model import VirtualStagingToolV2
+
+
+def predict(image, style, color_preference):
+    init_image = image.convert("RGB").resize((512, 512))
+    # mask = dict["mask"].convert("RGB").resize((512, 512))
+
+    vs_tool = VirtualStagingToolV2(diffusion_version="stabilityai/stable-diffusion-2-inpainting")
+    output_images, transparent_mask_image = vs_tool.virtual_stage(
+        image=init_image, style=style, color_preference=color_preference, number_images=1)
+    return output_images[0], transparent_mask_image, gr.update(visible=True), gr.update(visible=True), gr.update(visible=True)
+
+
+image_blocks = gr.Blocks()
+with image_blocks as demo:
+    with gr.Group():
+        with gr.Box():
+            with gr.Row():
+                with gr.Column():
+                    image = gr.Image(source='upload', elem_id="image_upload",
+                                     type="pil", label="Upload",
+                                     ).style(height=400)
+                    with gr.Row(elem_id="prompt-container").style(mobile_collapse=False, equal_height=True):
+                        style = gr.Dropdown(
+                            ["Mordern", "Coastal", "French country"],
+                            label="Design theme", elem_id="input-color"
+                        )
+
+                        color_preference = gr.Textbox(placeholder='Enter color preference',
+                                                      label="Color preference", elem_id="input-color")
+                        btn = gr.Button("Inpaint!").style(
+                            margin=False,
+                            rounded=(False, True, True, False),
+                            full_width=False,
+                        )
+                with gr.Column():
+                    mask_image = gr.Image(label="Mask image", elem_id="mask-img").style(height=400)
+                    image_out = gr.Image(label="Output", elem_id="output-img").style(height=400)
+
+            btn.click(fn=predict, inputs=[image, style, color_preference], outputs=[image_out, mask_image])
+
+image_blocks.launch()

model.py

@@ -0,0 +1,147 @@
+from collections import defaultdict
+import matplotlib.pyplot as plt
+import matplotlib.patches as mpatches
+from matplotlib import cm
+
+from PIL import Image
+
+import torch
+from transformers import AutoImageProcessor, UperNetForSemanticSegmentation
+from transformers import SegformerFeatureExtractor, SegformerForSemanticSegmentation
+from diffusers import StableDiffusionInpaintPipeline
+
+
+class VirtualStagingToolV2():
+
+    def __init__(self,
+                 segmentation_version='openmmlab/upernet-convnext-tiny',
+                 diffusion_version="stabilityai/stable-diffusion-2-inpainting"
+                 ):
+
+        self.segmentation_version = segmentation_version
+        self.diffusion_version = diffusion_version
+
+        self.feature_extractor = AutoImageProcessor.from_pretrained(self.segmentation_version)
+        self.segmentation_model = UperNetForSemanticSegmentation.from_pretrained(self.segmentation_version)
+
+        self.diffution_pipeline = StableDiffusionInpaintPipeline.from_pretrained(
+            self.diffusion_version,
+            torch_dtype=torch.float32,
+        )
+        self.diffution_pipeline = self.diffution_pipeline.to("cpu")
+
+    def _predict(self, image):
+        inputs = self.feature_extractor(images=image, return_tensors="pt")
+        outputs = self.segmentation_model(**inputs)
+        prediction = \
+        self.feature_extractor.post_process_semantic_segmentation(outputs, target_sizes=[image.size[::-1]])[0]
+        return prediction
+
+    def _save_mask(self, img, prediction_array, mask_items=[]):
+        mask = np.zeros_like(prediction_array, dtype=np.uint8)
+
+        mask[np.isin(prediction_array, mask_items)] = 0
+        mask[~np.isin(prediction_array, mask_items)] = 255
+
+        #     # # Create a PIL Image object from the mask
+        mask_image = Image.fromarray(mask, mode='L')
+        # display(mask_image)
+
+        # mask_image = mask_image.resize((512, 512))
+        # mask_image.save(".tmp/mask_1.png", "PNG")
+        # img = img.resize((512, 512))
+        # img.save(".tmp/input_1.png", "PNG")
+        return mask_image
+
+    def _save_transparent_mask(self, img, prediction_array, mask_items=[]):
+        mask = np.array(img)
+        mask[~np.isin(prediction_array, mask_items), :] = 255
+        mask_image = Image.fromarray(mask).convert('RGBA')
+
+        # Set the transparency of the pixels corresponding to object 1 to 0 (fully transparent)
+        mask_data = mask_image.getdata()
+        mask_data = [(r, g, b, 0) if r == 255 else (r, g, b, 255) for (r, g, b, a) in mask_data]
+        mask_image.putdata(mask_data)
+
+        return mask_image
+
+    def get_mask(self, image_path=None, image=None):
+        if image_path:
+            image = Image.open(image_path)
+        else:
+            if not image:
+                raise ValueError("no image provided")
+
+        # display(image)
+        prediction = self._predict(image)
+
+        label_ids = np.unique(prediction)
+
+        mask_items = [0, 3, 5, 8, 14]
+
+        if 1 in label_ids or 25 in label_ids:
+            mask_items = [1, 2, 4, 25, 32]
+            room = 'backyard'
+        elif 73 in label_ids or 50 in label_ids or 61 in label_ids:
+            mask_items = [0, 3, 5, 8, 14, 50, 61, 71, 118, 124, 129
+                          ]
+            room = 'kitchen'
+        elif 37 in label_ids or 65 in label_ids or (27 in label_ids and 47 in label_ids and 70 in label_ids):
+            mask_items = [0, 3, 5, 8, 14, 27, 65]
+            room = 'bathroom'
+        elif 7 in label_ids:
+            room = 'bedroom'
+        elif 23 in label_ids or 49 in label_ids:
+            room = 'living room'
+
+        label_ids_without_mask = [i for i in label_ids if i not in mask_items]
+
+        items = [self.segmentation_model.config.id2label[i] for i in label_ids_without_mask]
+
+        mask_image = self._save_mask(image, prediction, mask_items)
+        transparent_mask_image = self._save_transparent_mask(image, prediction, mask_items)
+        return mask_image, transparent_mask_image, image, items, room
+
+    def _edit_image(self, init_image, mask_image, prompt,  # height, width,
+                    number_images=1):
+
+        init_image = init_image.resize((512, 512)).convert("RGB")
+        mask_image = mask_image.resize((512, 512)).convert("RGB")
+
+        display(init_image)
+        display(mask_image)
+
+        output_images = self.diffution_pipeline(
+            prompt=prompt, image=init_image, mask_image=mask_image,
+            # width=width, height=height,
+            num_images_per_prompt=number_images).images
+        # display(output_image)
+        return output_images
+
+    def virtual_stage(self, image_path=None, image=None, style=None, color_preference=None, number_images=1):
+        mask_image, transparent_mask_image, init_image, items, room = self.get_mask(image_path, image)
+        if not style:
+            raise ValueError('style not provided.')
+        if not color_preference:
+            raise ValueError('color_preference not provided.')
+
+        if room == 'kitchen':
+            items = [i for i in items if i in ['kitchen island', 'cabinet', 'shelf', 'counter', 'countertop', 'stool']]
+        elif room == 'bedroom':
+            items = [i for i in items if i in ['bed', 'table', 'chest of drawers', 'desk', 'armchair', 'wardrobe']]
+        elif room == 'bathroom':
+            items = [i for i in items if
+                     i in ['shower', 'bathtub', 'chest of drawers', 'counter', 'countertop', 'sink']]
+
+        items = ', '.join(items)
+        prompt = f'{items}, high resolution, in the {style} style {room} in {color_preference}'
+        print(prompt)
+
+        output_images = self._edit_image(init_image, mask_image, prompt, number_images)
+
+        final_output_images = []
+        for output_image in output_images:
+            display(output_image)
+            output_image = output_image.resize(init_image.size)
+            final_output_images.append(output_image)
+        return final_output_images, transparent_mask_image

requirements.txt

@@ -0,0 +1,6 @@
+transformers==4.29.0
+torch==1.11.0
+diffusers==0.16.1
+accelerate==0.19.0
+matplotlib==3.6.2
+pillow==9.2.0