106 lines
7.5 KiB
XML
106 lines
7.5 KiB
XML
<svg width="700" height="280" xmlns="http://www.w3.org/2000/svg">
|
||
<defs>
|
||
<marker id="dl-arrow" markerWidth="7" markerHeight="5" refX="7" refY="2.5" orient="auto">
|
||
<polygon points="0 0, 7 2.5, 0 5" fill="#555"/>
|
||
</marker>
|
||
</defs>
|
||
<text x="350" y="22" fill="#333" font-size="14" font-weight="bold" text-anchor="middle">DeepLab: Atrous Spatial Pyramid Pooling (ASPP)</text>
|
||
|
||
<!-- Left: Atrous convolution illustration -->
|
||
<text x="130" y="48" fill="#333" font-size="11" font-weight="bold" text-anchor="middle">Atrous (Dilated) Convolution</text>
|
||
|
||
<!-- Rate 1 (standard 3x3) -->
|
||
<text x="60" y="70" fill="#3498db" font-size="9" text-anchor="middle" font-weight="bold">rate=1</text>
|
||
<g transform="translate(25, 75)">
|
||
<rect x="0" y="0" width="70" height="70" fill="#eee" stroke="#ccc" stroke-width="0.5"/>
|
||
<!-- 5x5 grid, centre 3x3 filled -->
|
||
<rect x="14" y="14" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="28" y="14" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="42" y="14" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="14" y="28" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="28" y="28" width="14" height="14" fill="#3498db" opacity="0.6" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="42" y="28" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="14" y="42" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="28" y="42" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
<rect x="42" y="42" width="14" height="14" fill="#3498db" opacity="0.4" stroke="#3498db" stroke-width="1"/>
|
||
</g>
|
||
<text x="60" y="158" fill="#3498db" font-size="8" text-anchor="middle">RF: 3×3</text>
|
||
|
||
<!-- Rate 2 -->
|
||
<text x="170" y="70" fill="#e74c3c" font-size="9" text-anchor="middle" font-weight="bold">rate=2</text>
|
||
<g transform="translate(135, 75)">
|
||
<rect x="0" y="0" width="70" height="70" fill="#eee" stroke="#ccc" stroke-width="0.5"/>
|
||
<!-- Dilated: skip one -->
|
||
<rect x="0" y="0" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="28" y="0" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="56" y="0" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="0" y="28" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="28" y="28" width="14" height="14" fill="#e74c3c" opacity="0.6" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="56" y="28" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="0" y="56" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="28" y="56" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
<rect x="56" y="56" width="14" height="14" fill="#e74c3c" opacity="0.4" stroke="#e74c3c" stroke-width="1"/>
|
||
</g>
|
||
<text x="170" y="158" fill="#e74c3c" font-size="8" text-anchor="middle">RF: 5×5, 9 params</text>
|
||
|
||
<!-- ASPP module (right side) -->
|
||
<text x="475" y="48" fill="#333" font-size="11" font-weight="bold" text-anchor="middle">ASPP Module</text>
|
||
|
||
<!-- Input feature map -->
|
||
<rect x="290" y="68" width="60" height="60" rx="3" fill="#ecf0f1" stroke="#999" stroke-width="1.5"/>
|
||
<text x="320" y="102" fill="#333" font-size="8" text-anchor="middle">Features</text>
|
||
|
||
<!-- Parallel branches -->
|
||
<!-- Branch 1: 1x1 conv -->
|
||
<line x1="350" y1="78" x2="395" y2="68" stroke="#555" stroke-width="1" marker-end="url(#dl-arrow)"/>
|
||
<rect x="400" y="55" width="70" height="25" rx="4" fill="#3498db" opacity="0.15" stroke="#3498db" stroke-width="1.5"/>
|
||
<text x="435" y="72" fill="#3498db" font-size="8" text-anchor="middle" font-weight="bold">1×1 (r=1)</text>
|
||
|
||
<!-- Branch 2: 3x3 rate 6 -->
|
||
<line x1="350" y1="88" x2="395" y2="88" stroke="#555" stroke-width="1" marker-end="url(#dl-arrow)"/>
|
||
<rect x="400" y="82" width="70" height="25" rx="4" fill="#e74c3c" opacity="0.15" stroke="#e74c3c" stroke-width="1.5"/>
|
||
<text x="435" y="99" fill="#e74c3c" font-size="8" text-anchor="middle" font-weight="bold">3×3 (r=6)</text>
|
||
|
||
<!-- Branch 3: 3x3 rate 12 -->
|
||
<line x1="350" y1="98" x2="395" y2="108" stroke="#555" stroke-width="1" marker-end="url(#dl-arrow)"/>
|
||
<rect x="400" y="109" width="70" height="25" rx="4" fill="#27ae60" opacity="0.15" stroke="#27ae60" stroke-width="1.5"/>
|
||
<text x="435" y="126" fill="#27ae60" font-size="8" text-anchor="middle" font-weight="bold">3×3 (r=12)</text>
|
||
|
||
<!-- Branch 4: 3x3 rate 18 -->
|
||
<line x1="350" y1="108" x2="395" y2="128" stroke="#555" stroke-width="1" marker-end="url(#dl-arrow)"/>
|
||
<rect x="400" y="136" width="70" height="25" rx="4" fill="#f39c12" opacity="0.15" stroke="#f39c12" stroke-width="1.5"/>
|
||
<text x="435" y="153" fill="#f39c12" font-size="8" text-anchor="middle" font-weight="bold">3×3 (r=18)</text>
|
||
|
||
<!-- Branch 5: Global Average Pool -->
|
||
<line x1="350" y1="118" x2="395" y2="148" stroke="#555" stroke-width="1" marker-end="url(#dl-arrow)"/>
|
||
<rect x="400" y="163" width="70" height="25" rx="4" fill="#9b59b6" opacity="0.15" stroke="#9b59b6" stroke-width="1.5"/>
|
||
<text x="435" y="180" fill="#9b59b6" font-size="8" text-anchor="middle" font-weight="bold">GAP + 1×1</text>
|
||
|
||
<!-- All branches merge -->
|
||
<line x1="470" y1="68" x2="510" y2="120" stroke="#555" stroke-width="1"/>
|
||
<line x1="470" y1="95" x2="510" y2="120" stroke="#555" stroke-width="1"/>
|
||
<line x1="470" y1="122" x2="510" y2="120" stroke="#555" stroke-width="1"/>
|
||
<line x1="470" y1="149" x2="510" y2="120" stroke="#555" stroke-width="1"/>
|
||
<line x1="470" y1="176" x2="510" y2="120" stroke="#555" stroke-width="1"/>
|
||
|
||
<!-- Concat -->
|
||
<rect x="515" y="105" width="55" height="30" rx="4" fill="#333" opacity="0.08" stroke="#333" stroke-width="1.5"/>
|
||
<text x="543" y="124" fill="#333" font-size="8" text-anchor="middle" font-weight="bold">Concat</text>
|
||
|
||
<!-- 1x1 conv fusion -->
|
||
<line x1="570" y1="120" x2="595" y2="120" stroke="#555" stroke-width="1.2" marker-end="url(#dl-arrow)"/>
|
||
<rect x="600" y="105" width="55" height="30" rx="4" fill="#333" opacity="0.08" stroke="#333" stroke-width="1.5"/>
|
||
<text x="628" y="124" fill="#333" font-size="8" text-anchor="middle" font-weight="bold">1×1 conv</text>
|
||
|
||
<!-- Output -->
|
||
<line x1="655" y1="120" x2="680" y2="120" stroke="#555" stroke-width="1.2" marker-end="url(#dl-arrow)"/>
|
||
<text x="690" y="124" fill="#333" font-size="8" text-anchor="start">Out</text>
|
||
|
||
<!-- Bottom notes -->
|
||
<rect x="50" y="200" width="600" height="70" rx="6" fill="#f5f5f5" stroke="#333" stroke-width="1"/>
|
||
<text x="350" y="220" fill="#333" font-size="10" text-anchor="middle" font-weight="bold">How Atrous Convolution Works</text>
|
||
<text x="350" y="236" fill="#666" font-size="9" text-anchor="middle">Standard 3×3 filter with gaps of (rate−1) between elements. Rate r gives receptive field (2r+1)×(2r+1).</text>
|
||
<text x="350" y="252" fill="#666" font-size="9" text-anchor="middle">ASPP applies multiple rates in parallel to capture context at multiple scales — like Inception but with dilation.</text>
|
||
<text x="350" y="264" fill="#666" font-size="9" text-anchor="middle">Global average pooling branch captures image-level context (what's the overall scene?).</text>
|
||
</svg>
|