include visualizations

README.md

@@ -8,11 +8,14 @@ size_categories:
 ---
 # DeepFurniture Dataset
 
-A large-scale dataset for furniture understanding, featuring **photo-realistic rendered indoor scenes** with **high-quality 3D furniture models**. The dataset contains about 24k indoor images, 170k furniture instances, and 20k unique furniture identities, all rendered by the leading industry-level rendering engines in [Kujiale](https://coohom.com).
-
 This dataset is introduced in our paper:
 [Furnishing Your Room by What You See: An End-to-End Furniture Set Retrieval Framework with Rich Annotated Benchmark Dataset](https://arxiv.org/abs/1911.09299)
 
+<img src="visualizations/overview.png" width="100%"/>
+
+A large-scale dataset for furniture understanding, featuring **photo-realistic rendered indoor scenes** with **high-quality 3D furniture models**. The dataset contains about 24k indoor images, 170k furniture instances, and 20k unique furniture identities, all rendered by the leading industry-level rendering engines in [Kujiale](https://coohom.com).
+
+
 ## Key Features
 
 - **Photo-Realistic Rendering**: All indoor scenes are rendered using professional rendering engines, providing realistic lighting, shadows, and textures
@@ -34,31 +37,21 @@ DeepFurniture provides hierarchical annotations at three levels:
 - Categories: 11 furniture types
 - Style tags: 11 different styles
 
-### Categories
-1. Cabinet/Shelf
-2. Table
-3. Chair/Stool
-4. Lamp
-5. Door
-6. Bed
-7. Sofa
-8. Plant
-9. Decoration
-10. Curtain
-11. Home Appliance
-
-### Style Tags
-1. Modern
-2. Country
-3. European/American
-4. Chinese
-5. Japanese
-6. Mediterranean
-7. Southeast-Asian
-8. Nordic
-9. Industrial
-10. Eclectic
-11. Other
+## Dataset Visualization
+
+You can view example visualizations of the dataset [here](visualizations/index.html). These examples show:
+- Scene images with instance annotations
+- Depth maps
+- Furniture instance details
+
+## Benchmarks
+
+This dataset supports three main benchmarks:
+1. Furniture Detection/Segmentation
+2. Furniture Instance Retrieval 
+3. Furniture Retrieval
+
+For benchmark details and baselines, please refer to our paper.
 
 ## Dataset Structure
 
@@ -85,9 +78,13 @@ data/
 # Clone the repository
 git lfs install  # Make sure Git LFS is installed
 git clone https://huggingface.co/datasets/byliu/DeepFurniture
+```
 
-# [optional] Extract the dataset (or load the dataset using the compressed files, TBD) 
-./uncompress_dataset.sh -s ./DeepFurniture/data -t ./regular_dataset
+[optional] Uncompress the dataset. The current dataset loader is only available for uncompressed assets. So, if you want to use the provided dataset loader, you'll need the ucnompress the dataset firstly.
+The dataset loader for compressed assets is TBD.
+```
+cd DeepFurniture
+bash uncompress_dataset.sh -s data -t uncompressed_data
 ```
 
 ### 2. Data Format
@@ -132,7 +129,7 @@ git clone https://huggingface.co/datasets/byliu/DeepFurniture
 from deepfurniture import DeepFurnitureDataset
 
 # Initialize dataset
-dataset = DeepFurnitureDataset("path/to/regular_dataset")
+dataset = DeepFurnitureDataset("path/to/uncompressed_data")
 
 # Access a scene
 scene = dataset[0]
@@ -145,14 +142,11 @@ for instance in scene['instances']:
     print(f"Style(s): {instance['style_names']}")
 ```
 
-## Benchmarks
-
-This dataset supports three main benchmarks:
-1. Furniture Detection/Segmentation
-2. Furniture Instance Retrieval 
-3. Furniture Retrieval
+### 4. To visualize each indoor scene
+```
+python visualize_html.py --dataset ./uncompressed_data --scene_idx 101 --output scene_101.html
+```
 
-For benchmark details and baselines, please refer to our paper.
 
 
 ## License

data/metadata/categories.json

@@ -1,3 +1,13 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:564f823a86a8501c6d02ec39dc155df85bb5d875bd155cff9dc3f1fe3c375555
-size 205
+{
+  "1": "cabinet#shelf",
+  "2": "table",
+  "3": "chair#stool",
+  "4": "lamp",
+  "5": "door",
+  "6": "bed",
+  "7": "sofa",
+  "8": "plant",
+  "9": "decoration",
+  "10": "curtain",
+  "11": "home-appliance"
+}

data/metadata/dataset_info.json

@@ -1,3 +1,12 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:52882b28624b44aa66b2d3e7c41ddd38514f312ee3cb61b781ca13f174a1f480
-size 312
+{
+  "name": "DeepFurniture",
+  "version": "1.0.0",
+  "description": "A large-scale dataset for furniture understanding with rich annotations",
+  "citation": "Add citation here",
+  "license": "Add license here",
+  "statistics": {
+    "num_scenes": 24182,
+    "num_furnitures": 24742,
+    "num_queries": 7264
+  }
+}

data/metadata/styles.json

@@ -1,3 +1,13 @@
-version https://git-lfs.github.com/spec/v1
-oid sha256:872861a40d355a5d378bc7e2909878633d030f2ec3adc70785528a2f566ecee3
-size 226
+{
+  "1": "modern",
+  "2": "country",
+  "3": "European#American",
+  "4": "Chinese",
+  "5": "Japanese",
+  "6": "Mediterranean",
+  "7": "Southeast-Asian",
+  "8": "Nordic",
+  "9": "Industrial",
+  "10": "electic",
+  "11": "other"
+}

uncompress_dataset.sh

@@ -2,7 +2,7 @@
 
 # Usage function
 usage() {
-    echo "Usage: $0 -s SOURCE_DIR -t TARGET_DIR [-c CHUNK_TYPE] [-h]"
+    echo "Usage: $0 -s SOURCE_DIR -t TARGET_DIR [-c CHUNK_TYPE] [-m MAX_FILES] [-h]"
     echo "Uncompress chunked DeepFurniture dataset"
     echo ""
     echo "Required arguments:"
@@ -12,16 +12,18 @@ usage() {
     echo "Optional arguments:"
     echo "  -c CHUNK_TYPE   Specific chunk type to process (scenes, furnitures, queries)"
     echo "                  If not specified, all chunk types will be processed"
+    echo "  -m MAX_FILES    Maximum number of files to process per type (default: process all)"
     echo "  -h             Show this help message"
     exit 1
 }
 
 # Process command line arguments
-while getopts "s:t:c:h" opt; do
+while getopts "s:t:c:m:h" opt; do
     case $opt in
         s) SOURCE_DIR="$OPTARG";;
         t) TARGET_DIR="$OPTARG";;
         c) CHUNK_TYPE="$OPTARG";;
+        m) MAX_FILES="$OPTARG";;
         h) usage;;
         ?) usage;;
     esac
@@ -39,6 +41,15 @@ if [ ! -d "$SOURCE_DIR" ]; then
     exit 1
 fi
 
+# Validate MAX_FILES if provided
+if [ -n "$MAX_FILES" ]; then
+    if ! [[ "$MAX_FILES" =~ ^[0-9]+$ ]]; then
+        echo "Error: MAX_FILES must be a positive integer"
+        exit 1
+    fi
+    echo "Will process maximum $MAX_FILES files per type"
+fi
+
 # Create target directory structure
 mkdir -p "$TARGET_DIR"/{metadata,scenes,furnitures,queries}
 
@@ -62,21 +73,32 @@ process_chunks() {
     if [ ! -d "$src_dir" ]; then
         echo "Warning: Directory not found: $src_dir"
         return
-    }
+    fi
+    
+    # Get list of chunks and sort them
+    chunks=($(ls -v "$src_dir"/*.tar.gz 2>/dev/null))
+    total_chunks=${#chunks[@]}
     
-    # Count total chunks for progress
-    total_chunks=$(ls "$src_dir"/*.tar.gz 2>/dev/null | wc -l)
     if [ "$total_chunks" -eq 0 ]; then
         echo "No chunks found in $src_dir"
         return
-    }
+    fi
+    
+    # Determine how many chunks to process based on MAX_FILES
+    files_per_chunk=1000  # Default files per chunk based on dataset structure
+    if [ -n "$MAX_FILES" ]; then
+        chunks_needed=$(( (MAX_FILES + files_per_chunk - 1) / files_per_chunk ))
+        if [ "$chunks_needed" -lt "$total_chunks" ]; then
+            total_chunks=$chunks_needed
+            echo "Limiting to $total_chunks chunks ($MAX_FILES files) for $type"
+        fi
+    fi
     
-    # Process each chunk
-    current=0
-    for chunk in "$src_dir"/*.tar.gz; do
-        current=$((current + 1))
+    # Process chunks
+    for ((i = 0; i < total_chunks; i++)); do
+        chunk="${chunks[$i]}"
         chunk_name=$(basename "$chunk")
-        printf "Extracting %s (%d/%d)..." "$chunk_name" "$current" "$total_chunks"
+        printf "Extracting %s (%d/%d)..." "$chunk_name" $((i + 1)) "$total_chunks"
         
         if tar -xzf "$chunk" -C "$target_dir" 2>/dev/null; then
             echo " done"
@@ -84,6 +106,21 @@ process_chunks() {
             echo " failed"
             echo "Warning: Failed to extract $chunk_name"
         fi
+        
+        # If this is the last chunk and we have MAX_FILES set,
+        # we might need to remove excess files
+        if [ -n "$MAX_FILES" ] && [ "$i" -eq "$((total_chunks - 1))" ]; then
+            # Calculate how many files we should have
+            local expected_total=$MAX_FILES
+            local current_total=$(ls "$target_dir" | wc -l)
+            
+            if [ "$current_total" -gt "$expected_total" ]; then
+                echo "Trimming excess files to meet MAX_FILES limit..."
+                # Remove excess files (keeping the first MAX_FILES files)
+                ls "$target_dir" | sort | tail -n+"$((expected_total + 1))" | \
+                    xargs -I {} rm -rf "$target_dir/{}"
+            fi
+        fi
     done
 }
 
@@ -112,8 +149,10 @@ echo -e "\nValidating extracted files..."
 # Check scenes
 if [ -z "$CHUNK_TYPE" ] || [ "$CHUNK_TYPE" = "scenes" ]; then
     missing_files=0
+    total_scenes=0
     for scene_dir in "$TARGET_DIR"/scenes/*; do
         if [ -d "$scene_dir" ]; then
+            total_scenes=$((total_scenes + 1))
             for required in "image.jpg" "annotation.json"; do
                 if [ ! -f "$scene_dir/$required" ]; then
                     echo "Warning: Missing $required in $(basename "$scene_dir")"
@@ -122,8 +161,17 @@ if [ -z "$CHUNK_TYPE" ] || [ "$CHUNK_TYPE" = "scenes" ]; then
             done
         fi
     done
-    echo "Scene validation complete. Missing files: $missing_files"
+    echo "Scene validation complete. Processed $total_scenes scenes. Missing files: $missing_files"
 fi
 
+# Print final statistics
+echo -e "\nExtraction Summary:"
+for type in scenes furnitures queries; do
+    if [ -z "$CHUNK_TYPE" ] || [ "$CHUNK_TYPE" = "$type" ]; then
+        file_count=$(find "$TARGET_DIR/$type" -type f | wc -l)
+        echo "$type: $file_count files"
+    fi
+done
+
 echo "Dataset uncompression completed!"
 echo "Output directory: $TARGET_DIR"

visualizations/index.html

@@ -0,0 +1,78 @@
+
+    <!DOCTYPE html>
+    <html>
+    <head>
+        <title>DeepFurniture Dataset Visualizations</title>
+        <style>
+            body { 
+                font-family: Arial; 
+                max-width: 1200px; 
+                margin: 0 auto; 
+                padding: 20px;
+            }
+            .example-grid {
+                display: grid;
+                grid-template-columns: repeat(auto-fit, minmax(300px, 1fr));
+                gap: 20px;
+                margin: 20px 0;
+            }
+            .example-card {
+                border: 1px solid #ddd;
+                border-radius: 8px;
+                padding: 15px;
+                text-decoration: none;
+                color: inherit;
+                transition: transform 0.2s;
+            }
+            .example-card:hover {
+                transform: translateY(-5px);
+                box-shadow: 0 4px 8px rgba(0,0,0,0.1);
+            }
+            h1 { text-align: center; }
+        </style>
+    </head>
+    <body>
+        <h1>DeepFurniture Dataset Visualizations</h1>
+        <p>This page shows example visualizations from the DeepFurniture dataset. Click on any example to view the full visualization.</p>
+        
+        <div class="example-grid">
+    
+            <a href="example_0.html" class="example-card">
+                <h3>Scene 1</h3>
+                <p>Scene ID: DVD3MBOCEJLMIK6A573WKUY8</p>
+                <p>Number of instances: 11</p>
+                <p>Click to view details →</p>
+            </a>
+        
+            <a href="example_1.html" class="example-card">
+                <h3>Scene 2</h3>
+                <p>Scene ID: DVD3MHYMEJI4OKYQBT3WKSY8</p>
+                <p>Number of instances: 4</p>
+                <p>Click to view details →</p>
+            </a>
+        
+            <a href="example_2.html" class="example-card">
+                <h3>Scene 3</h3>
+                <p>Scene ID: DVD5FHHPEJJ4ESRZRX3WKUQ8</p>
+                <p>Number of instances: 8</p>
+                <p>Click to view details →</p>
+            </a>
+        
+            <a href="example_3.html" class="example-card">
+                <h3>Scene 4</h3>
+                <p>Scene ID: DVD7IYG2EJLMOK6VYL3WKVY8</p>
+                <p>Number of instances: 15</p>
+                <p>Click to view details →</p>
+            </a>
+        
+            <a href="example_4.html" class="example-card">
+                <h3>Scene 5</h3>
+                <p>Scene ID: DVDZU2DAEJI4OK6KYT3WKRA8</p>
+                <p>Number of instances: 6</p>
+                <p>Click to view details →</p>
+            </a>
+        
+        </div>
+    </body>
+    </html>
+    

visualize_html.py

@@ -0,0 +1,292 @@
+import numpy as np
+from PIL import Image, ImageDraw, ImageFont
+import base64
+import io
+from deepfurniture import DeepFurnitureDataset
+from pycocotools import mask as mask_utils
+
+def save_image_base64(image):
+    """Convert PIL image to base64 string."""
+    buffered = io.BytesIO()
+    image.save(buffered, format="JPEG", quality=90)
+    return base64.b64encode(buffered.getvalue()).decode()
+
+
+def create_instance_visualization(scene_data):
+    """Create combined instance visualization with both masks and bboxes."""
+    image = scene_data['image']
+    instances = scene_data['instances']
+    
+    # Image dimensions for boundary checking
+    img_width, img_height = image.size
+    
+    # Start with image at half opacity
+    vis_img = np.array(image, dtype=np.float32) * 0.5
+    
+    # Get all segmentations
+    segmentations = []
+    for inst in instances:
+        if inst['segmentation']:
+            rle = {
+                'counts': inst['segmentation'],
+                'size': [img_height, img_width]
+            }
+            segmentations.append(rle)
+    
+    # Create color map for instances with distinct colors
+    colors = np.array([
+        [0.9, 0.1, 0.1],  # Red
+        [0.1, 0.9, 0.1],  # Green
+        [0.1, 0.1, 0.9],  # Blue
+        [0.9, 0.9, 0.1],  # Yellow
+        [0.9, 0.1, 0.9],  # Magenta
+        [0.1, 0.9, 0.9],  # Cyan
+        [0.9, 0.5, 0.1],  # Orange
+        [0.5, 0.9, 0.1],  # Lime
+        [0.5, 0.1, 0.9],  # Purple
+    ])
+    colors = np.tile(colors, (len(instances) // len(colors) + 1, 1))[:len(instances)]
+    
+    # Draw instance masks with higher opacity
+    if segmentations:
+        if isinstance(segmentations[0]['counts'], (list, tuple)):
+            segmentations = mask_utils.frPyObjects(
+                segmentations, img_height, img_width
+            )
+        masks = mask_utils.decode(segmentations)
+        
+        for idx in range(masks.shape[2]):
+            color = colors[idx]
+            mask = masks[:, :, idx]
+            for c in range(3):
+                vis_img[:, :, c] += mask * np.array(image)[:, :, c] * 0.7 * color[c]
+    
+    # Convert to PIL for drawing bounding boxes
+    vis_img = Image.fromarray(np.uint8(np.clip(vis_img, 0, 255)))
+    draw = ImageDraw.Draw(vis_img)
+    
+    # Try to load a font for better text rendering
+    try:
+        font = ImageFont.truetype("/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf", 20)
+    except:
+        try:
+            font = ImageFont.truetype("/System/Library/Fonts/Helvetica.ttc", 20)
+        except:
+            font = ImageFont.load_default()
+    
+    # Constants for text and box drawing
+    text_padding = 4
+    text_height = 24
+    text_width = 200
+    corner_length = 20
+    
+    # Draw bounding boxes with labels
+    for idx, (instance, color) in enumerate(zip(instances, colors)):
+        bbox = instance['bounding_box']
+        color_tuple = tuple(int(c * 255) for c in color)
+        
+        # Calculate label
+        furniture_id = instance['identity_id']
+        category = instance['category_name']
+        label = f"{category} ({furniture_id})"
+        
+        # Draw bbox with double lines for better visibility
+        for offset in [2, 1]:
+            draw.rectangle([
+                max(0, bbox['xmin'] - offset),
+                max(0, bbox['ymin'] - offset),
+                min(img_width - 1, bbox['xmax'] + offset),
+                min(img_height - 1, bbox['ymax'] + offset)
+            ], outline=color_tuple, width=2)
+        
+        # Determine text position (handle boundary cases)
+        # First try above the bbox
+        text_y = bbox['ymin'] - text_height - text_padding
+        if text_y < 0:  # If no space above, try below
+            text_y = bbox['ymax'] + text_padding
+            
+        # Handle x position
+        text_x = bbox['xmin']
+        # If text would go beyond right edge, align to right edge
+        if text_x + text_width > img_width:
+            text_x = max(0, img_width - text_width)
+        
+        # Draw background for text
+        text_pos = (text_x, text_y)
+        draw.rectangle([
+            text_pos[0] - 2,
+            text_pos[1] - 2,
+            min(img_width - 1, text_pos[0] + text_width),
+            min(img_height - 1, text_pos[1] + text_height)
+        ], fill='black')
+        
+        # Draw text
+        draw.text(text_pos, label, fill=color_tuple, font=font)
+        
+        # Add corner markers with boundary checking
+        corners = [
+            (bbox['xmin'], bbox['ymin']),  # Top-left
+            (bbox['xmax'], bbox['ymin']),  # Top-right
+            (bbox['xmin'], bbox['ymax']),  # Bottom-left
+            (bbox['xmax'], bbox['ymax'])   # Bottom-right
+        ]
+        
+        for x, y in corners:
+            # Ensure corner markers stay within image bounds
+            # Horizontal lines
+            x1 = max(0, x - corner_length)
+            x2 = min(img_width - 1, x + corner_length)
+            draw.line([(x1, y), (x2, y)], fill=color_tuple, width=3)
+            
+            # Vertical lines
+            y1 = max(0, y - corner_length)
+            y2 = min(img_height - 1, y + corner_length)
+            draw.line([(x, y1), (x, y2)], fill=color_tuple, width=3)
+    
+    return vis_img
+
+
+def process_depth_map(depth_image):
+    """Process depth map for better visualization.
+    
+    Args:
+        depth_image: PIL Image of depth map
+    Returns:
+        Processed depth map as PIL Image
+    """
+    # Convert to numpy array
+    depth = np.array(depth_image)
+    
+    # Normalize depth to 0-1 range
+    if depth.max() > depth.min():
+        depth = (depth - depth.min()) / (depth.max() - depth.min())
+    
+    # Apply colormap (viridis-like)
+    colored_depth = np.zeros((*depth.shape, 3))
+    colored_depth[..., 0] = (1 - depth) * 0.4  # Red channel
+    colored_depth[..., 1] = np.abs(depth - 0.5) * 0.8  # Green channel
+    colored_depth[..., 2] = depth * 0.8  # Blue channel
+    
+    # Convert to uint8 and then to PIL
+    colored_depth = (colored_depth * 255).astype(np.uint8)
+    return Image.fromarray(colored_depth)
+
+
+def visualize_html(dataset, scene_idx, output_path='scene.html'):
+    """Generate HTML visualization for a scene."""
+    scene_data = dataset[scene_idx]
+    
+    # Create visualizations
+    instance_vis = create_instance_visualization(scene_data)
+    
+    depth_vis = None
+    if scene_data['depth']:
+        depth_vis = process_depth_map(scene_data['depth'])
+    # Get base64 encoded images
+    scene_img = save_image_base64(scene_data['image'])
+    instance_vis = save_image_base64(instance_vis)
+    depth_img = save_image_base64(depth_vis) if depth_vis else None
+    
+    # Create HTML with minimal CSS
+    html = f'''
+    <html>
+    <head>
+        <style>
+            body {{ font-family: Arial; margin: 20px; max-width: 2000px; margin: 0 auto; }}
+            .grid {{ display: grid; grid-template-columns: repeat(auto-fill, minmax(300px, 1fr)); gap: 20px; }}
+            .main-images {{ 
+                grid-template-columns: repeat(auto-fit, minmax(800px, 1fr)); 
+                margin: 20px 0;
+            }}
+            .card {{ 
+                border: 1px solid #ddd; 
+                padding: 15px; 
+                border-radius: 8px;
+                box-shadow: 0 2px 4px rgba(0,0,0,0.1);
+            }}
+            .card h3 {{ 
+                font-size: 20px;
+                margin-bottom: 15px;
+            }}
+            img {{ max-width: 100%; height: auto; }}
+            h1 {{ 
+                color: #333; 
+                font-size: 32px;
+                text-align: center;
+                margin: 30px 0;
+            }}
+            h2 {{ 
+                color: #333; 
+                font-size: 28px;
+                margin: 25px 0;
+            }}
+            .instance-info {{ 
+                color: #444; 
+                font-size: 16px;
+                line-height: 1.4;
+            }}
+            .main-images img {{
+                width: 100%;
+                object-fit: contain;
+                max-height: 800px;  /* Increased max height */
+            }}
+        </style>
+    </head>
+    <body>
+        <h1>Scene ID: {scene_data['scene_id']}</h1>
+        
+        <h2>Scene Visualizations</h2>
+        <div class="grid main-images">
+            <div class="card">
+                <h3>Original Scene</h3>
+                <img src="data:image/png;base64,{scene_img}">
+            </div>
+            <div class="card">
+                <h3>Instance Visualization (Masks + Bboxes)</h3>
+                <img src="data:image/png;base64,{instance_vis}">
+            </div>
+            {f'<div class="card"><h3>Depth Map</h3><img src="data:image/png;base64,{depth_img}"></div>' if depth_img else ''}
+        </div>
+        
+        <h2>Furniture Instances</h2>
+        <div class="grid">
+    '''
+    
+    # Add furniture previews
+    for instance in scene_data['instances']:
+        furniture_id = str(instance['identity_id'])
+        if furniture_id in scene_data['furniture_previews']:
+            preview = save_image_base64(scene_data['furniture_previews'][furniture_id])
+            bbox = instance['bounding_box']
+            
+            html += f'''
+            <div class="card">
+                <h3>{instance['category_name']} (ID: {furniture_id})</h3>
+                <div class="instance-info">
+                    <p>Style: {', '.join(instance['style_names'])}</p>
+                    <p>BBox: ({bbox['xmin']}, {bbox['ymin']}, {bbox['xmax']}, {bbox['ymax']})</p>
+                </div>
+                <img src="data:image/png;base64,{preview}">
+            </div>
+            '''
+    
+    html += '''
+        </div>
+    </body>
+    </html>
+    '''
+    
+    with open(output_path, 'w') as f:
+        f.write(html)
+    print(f"Visualization saved to {output_path}")
+
+if __name__ == '__main__':
+    import argparse
+    parser = argparse.ArgumentParser()
+    parser.add_argument('--dataset', required=True)
+    parser.add_argument('--scene_idx', type=int, required=True)
+    parser.add_argument('--output', default='scene.html')
+    args = parser.parse_args()
+
+    dataset = DeepFurnitureDataset(args.dataset)
+    visualize_html(dataset, args.scene_idx, args.output)