presentation/visuals/project_scope_board.svg

<svg xmlns="http://www.w3.org/2000/svg" width="1600" height="900" viewBox="0 0 1600 900" fill="none">
    <defs>
      <linearGradient id="bg-light" x1="0" y1="0" x2="1" y2="1">
        <stop offset="0%" stop-color="#FFF8EE"/>
        <stop offset="100%" stop-color="#F4F7FB"/>
      </linearGradient>
    </defs>
    <rect width="1600" height="900" fill="url(#bg-light)"/>
    <line x1="80" y1="0" x2="80" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="220" y1="0" x2="220" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="360" y1="0" x2="360" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="500" y1="0" x2="500" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="640" y1="0" x2="640" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="780" y1="0" x2="780" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="920" y1="0" x2="920" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="1060" y1="0" x2="1060" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="1200" y1="0" x2="1200" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="1340" y1="0" x2="1340" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="1480" y1="0" x2="1480" y2="900" stroke="#E7E1D5" stroke-opacity="0.55" /><line x1="0" y1="120" x2="1600" y2="120" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="240" x2="1600" y2="240" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="360" x2="1600" y2="360" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="480" x2="1600" y2="480" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="600" x2="1600" y2="600" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="720" x2="1600" y2="720" stroke="#E7E1D5" stroke-opacity="0.35" /><line x1="0" y1="840" x2="1600" y2="840" stroke="#E7E1D5" stroke-opacity="0.35" /><text x="110" y="118" font-family="Fraunces, Georgia, Times New Roman, serif" font-size="52" fill="#11203C" font-weight="700">What We Trained vs What We Reused</text><text x="110" y="164" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="22" fill="#4B587C">A slide-ready scope board for explaining the final project story clearly.</text><rect x="105" y="220" width="650" height="560" rx="30" fill="#FFFBF5" stroke="#D97B37" stroke-width="3"/><text x="141" y="284" font-family="Fraunces, Georgia, Times New Roman, serif" font-size="42" fill="#11203C" font-weight="700">Trained by our team</text><rect x="141" y="304" width="120" height="8" rx="4" fill="#D97B37"/><rect x="845" y="220" width="650" height="560" rx="30" fill="#F7FAFF" stroke="#457BFF" stroke-width="3"/><text x="881" y="284" font-family="Fraunces, Georgia, Times New Roman, serif" font-size="42" fill="#11203C" font-weight="700">Used without training</text><rect x="881" y="304" width="120" height="8" rx="4" fill="#457BFF"/><rect x="145" y="280" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="171" y="308" width="94" height="28" rx="14" fill="#D97B37"/><text x="187" y="328" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">TRAINED</text><text x="287" y="326" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">WLASL-100 landmark CNN</text><text x="287" y="358" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Primary report-scale experiment. Trained</text><text x="287" y="384" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">on MediaPipe landmark sequences.</text><rect x="145" y="445" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="171" y="473" width="94" height="28" rx="14" fill="#D97B37"/><text x="187" y="493" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">TRAINED</text><text x="287" y="491" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">WLASL-25 landmark CNN</text><text x="287" y="523" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Smaller vocabulary for the live webcam</text><text x="287" y="549" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">demo and HF Space stability.</text><rect x="145" y="610" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="171" y="638" width="94" height="28" rx="14" fill="#D97B37"/><text x="187" y="658" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">TRAINED</text><text x="287" y="656" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">Landmark Transformer</text><text x="287" y="688" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Attention-based extension. Kept as</text><text x="287" y="714" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">extra-credit / modern-method evidence.</text><rect x="885" y="280" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="911" y="308" width="94" height="28" rx="14" fill="#457BFF"/><text x="927" y="328" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">REUSED</text><text x="1027" y="326" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">MediaPipe Holistic</text><text x="1027" y="358" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Frozen feature extractor that turns each</text><text x="1027" y="384" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">frame into 225 landmark coordinates.</text><rect x="885" y="445" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="911" y="473" width="94" height="28" rx="14" fill="#457BFF"/><text x="927" y="493" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">REUSED</text><text x="1027" y="491" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">Qwen2.5-VL local model</text><text x="1027" y="523" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Pretrained VLM used only as a zero-shot</text><text x="1027" y="549" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">reranker over the CNN top-5.</text><rect x="885" y="610" width="570" height="118" rx="22" fill="#FFFFFF" stroke="#E7EAF2" stroke-width="2"/><rect x="911" y="638" width="94" height="28" rx="14" fill="#457BFF"/><text x="927" y="658" font-family="IBM Plex Mono, Cascadia Code, Consolas, monospace" font-size="14" fill="#FFFFFF" font-weight="700">REUSED</text><text x="1027" y="656" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="28" fill="#11203C" font-weight="700">Gradio + Hugging Face Space</text><text x="1027" y="688" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">Presentation UI and deployment layer for</text><text x="1027" y="714" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="20" fill="#4B587C">the class demo.</text><rect x="110" y="815" width="1385" height="55" rx="28" fill="#11203C" opacity="0.94"/><text x="138" y="857" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="17" fill="#F7F1E8" font-weight="700">ONE-SENTENCE TAKEAWAY</text><text x="138" y="897" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="22" fill="#F7F1E8">We trained the recognition models ourselves, but we</text><text x="138" y="927" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="22" fill="#F7F1E8">compared them against a frozen local VLM instead of</text><text x="138" y="957" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="22" fill="#F7F1E8">fine-tuning the VLM. That still satisfies the project</text><text x="138" y="987" font-family="Aptos, Segoe UI, Helvetica, Arial, sans-serif" font-size="22" fill="#F7F1E8">scope because the trained CNN is the main model and the</text></svg>