app.py

from __future__ import annotations

import os
from pathlib import Path

import gradio as gr

from leaderboard import (
    DATA_PATH,
    SORT_CHOICES,
    filter_leaderboard,
    load_leaderboard,
    metric_standings,
    model_choices,
    render_methodology,
    render_profile,
    render_summary,
    render_table,
)
from submission import submit_score


ROOT = Path(__file__).parent
OVERVIEW_IMAGE = ROOT / "assets" / "avbench_outline.png"
SUBMISSION_TEMPLATE = ROOT / "data" / "submission_template.csv"

LEADERBOARD = load_leaderboard(DATA_PATH)
STANDINGS = metric_standings(LEADERBOARD)


def update_leaderboard(component_type: str, query: str, sort_by: str, sort_order: str):
    view = filter_leaderboard(LEADERBOARD, component_type, query, sort_by, sort_order)
    return render_summary(LEADERBOARD, view), render_table(view, STANDINGS, sort_by, sort_order)


def update_profile(model: str):
    return render_profile(LEADERBOARD, model)


CSS = """
:root {
  --avgen-bg: #f6f7f9;
  --avgen-panel: #ffffff;
  --avgen-ink: #18202b;
  --avgen-muted: #5f6b7a;
  --avgen-line: #d9dee7;
  --avgen-blue: #1d5f9f;
  --avgen-teal: #087a73;
  --avgen-orange: #b45309;
  --avgen-green: #24784f;
}

.gradio-container {
  max-width: 1280px !important;
  margin: 0 auto !important;
  background: var(--avgen-bg) !important;
  color: var(--avgen-ink) !important;
}

.app-header {
  display: grid;
  grid-template-columns: minmax(0, 1fr) auto;
  gap: 24px;
  align-items: end;
  padding: 26px 0 12px;
  border-bottom: 1px solid var(--avgen-line);
}

.app-header h1 {
  margin: 0;
  color: #111827 !important;
  font-size: clamp(28px, 4vw, 44px);
  font-weight: 850;
  line-height: 1.05;
  letter-spacing: 0;
}

.app-header p {
  margin: 10px 0 0;
  max-width: 860px;
  color: #4b5563 !important;
  font-size: 16px;
}

.header-links {
  display: flex;
  flex-wrap: wrap;
  gap: 8px;
  justify-content: flex-end;
}

.header-links a {
  color: var(--avgen-blue);
  text-decoration: none;
  border: 1px solid var(--avgen-line);
  background: #fff;
  padding: 8px 11px;
  border-radius: 8px;
  font-weight: 650;
}

.summary-grid {
  display: grid;
  grid-template-columns: repeat(5, minmax(130px, 1fr));
  gap: 10px;
  margin: 14px 0;
}

.summary-card,
.method-card {
  border: 1px solid var(--avgen-line);
  background: var(--avgen-panel);
  border-radius: 8px;
  padding: 12px;
}

.summary-card span,
.profile-total span,
.eyebrow {
  display: block;
  color: #4f5d6e !important;
  font-size: 12px;
  text-transform: uppercase;
  letter-spacing: 0;
  font-weight: 700;
}

.summary-card strong {
  display: block;
  margin-top: 4px;
  color: #111827 !important;
  font-size: 26px;
  font-weight: 850;
  line-height: 1;
}

.summary-card small {
  display: block;
  margin-top: 6px;
  color: #4b5563 !important;
  min-height: 32px;
}

.table-shell {
  overflow-x: auto;
  border: 1px solid var(--avgen-line);
  border-radius: 8px;
  background: #fff;
}

.leaderboard-table {
  width: 100%;
  border-collapse: collapse;
  min-width: 1180px;
  font-size: 14px;
}

.leaderboard-table th,
.leaderboard-table td {
  padding: 11px 10px;
  border-bottom: 1px solid #edf0f5;
  vertical-align: middle;
}

.leaderboard-table tbody td {
  color: var(--avgen-ink) !important;
}

.leaderboard-table thead th {
  position: sticky;
  top: 0;
  background: #f1f4f8;
  color: #202936 !important;
  text-align: left;
  white-space: nowrap;
  z-index: 1;
}

.leaderboard-table thead th.sorted {
  background: #e6eef8;
  color: #111827 !important;
  box-shadow: inset 0 -2px 0 var(--avgen-blue);
}

.sort-indicator {
  display: inline-block;
  margin-left: 6px;
  color: var(--avgen-blue);
  font-weight: 850;
}

.leaderboard-table tbody tr:hover {
  background: #faf7f1;
}

.rank-cell {
  width: 62px;
  color: #4f5d6e !important;
  font-weight: 700;
}

.model-cell {
  min-width: 180px;
  color: #16202d !important;
  font-weight: 750;
}

.components-cell {
  min-width: 220px;
}

.metric-cell {
  text-align: right;
  color: #253142 !important;
  font-weight: 650;
  font-variant-numeric: tabular-nums;
}

.metric-cell.best {
  color: #0f6a43 !important;
  font-weight: 800;
  background: #edf8f2;
}

.metric-cell.second {
  color: #8a4b08 !important;
  font-weight: 750;
  background: #fff7e8;
}

.component-badge,
.type-badge {
  display: inline-flex;
  align-items: center;
  max-width: 100%;
  margin: 2px 4px 2px 0;
  padding: 3px 8px;
  border-radius: 999px;
  font-size: 12px;
  font-weight: 700;
  white-space: nowrap;
  border: 1px solid transparent;
}

.component-badge.proprietary,
.type-badge.proprietary {
  color: var(--avgen-orange);
  background: #fff4e5;
  border-color: #f2d4aa;
}

.component-badge.open,
.type-badge.opensource {
  color: var(--avgen-blue);
  background: #edf5ff;
  border-color: #c7dff8;
}

.component-badge.neutral,
.type-badge.mixed {
  color: var(--avgen-teal);
  background: #eaf7f5;
  border-color: #bde0dc;
}

.profile-panel {
  display: grid;
  grid-template-columns: minmax(0, 1fr) auto;
  gap: 16px;
  border: 1px solid var(--avgen-line);
  background: #fff;
  border-radius: 8px;
  padding: 18px;
}

.profile-panel h2 {
  margin: 2px 0 8px;
  font-size: 26px;
}

.profile-total {
  min-width: 140px;
  text-align: right;
}

.profile-total strong {
  display: block;
  font-size: 42px;
  line-height: 1;
}

.profile-grid {
  grid-column: 1 / -1;
  display: grid;
  grid-template-columns: repeat(3, minmax(0, 1fr));
  gap: 10px;
}

.profile-metric {
  border: 1px solid #edf0f5;
  border-radius: 8px;
  padding: 10px;
}

.profile-metric-head {
  display: flex;
  justify-content: space-between;
  gap: 10px;
  font-variant-numeric: tabular-nums;
}

.bar-track {
  height: 8px;
  margin: 9px 0 6px;
  border-radius: 999px;
  background: #e7ebf1;
  overflow: hidden;
}

.bar-fill {
  height: 100%;
  border-radius: inherit;
  background: linear-gradient(90deg, var(--avgen-teal), var(--avgen-green));
}

.profile-metric small,
.methodology p {
  color: var(--avgen-muted);
}

.method-grid {
  display: grid;
  grid-template-columns: repeat(3, minmax(0, 1fr));
  gap: 10px;
  margin-bottom: 12px;
}

.method-card h3 {
  margin: 0 0 4px;
  font-size: 16px;
}

.method-card strong {
  display: block;
  font-size: 30px;
  color: var(--avgen-blue);
}

.empty-state {
  border: 1px solid var(--avgen-line);
  background: #fff;
  border-radius: 8px;
  padding: 22px;
  color: var(--avgen-muted);
}

.overview-image img {
  border: 1px solid var(--avgen-line);
  border-radius: 8px;
  background: #fff;
}

.submission-copy {
  color: var(--avgen-muted);
  margin: 0 0 14px;
}

.submission-copy strong {
  color: var(--avgen-ink);
}

.status-card {
  border: 1px solid var(--avgen-line);
  border-radius: 8px;
  padding: 14px;
  background: #fff;
}

.status-card strong {
  display: block;
  margin-bottom: 4px;
}

.status-card p {
  margin: 4px 0 0;
}

.status-card.success {
  border-color: #b8dbc9;
  background: #f1faf5;
}

.status-card.error {
  border-color: #efc2bd;
  background: #fff3f1;
}

@media (max-width: 980px) {
  .app-header {
    grid-template-columns: 1fr;
  }

  .header-links {
    justify-content: flex-start;
  }

  .summary-grid,
  .profile-grid,
  .method-grid {
    grid-template-columns: 1fr 1fr;
  }
}

@media (max-width: 640px) {
  .summary-grid,
  .profile-grid,
  .method-grid {
    grid-template-columns: 1fr;
  }

  .profile-panel {
    grid-template-columns: 1fr;
  }

  .profile-total {
    text-align: left;
  }
}
"""


HEADER = """
<div class="app-header">
  <div>
    <h1>AVGen-Bench Leaderboard</h1>
    <p>
      A leaderboard for multi-granular evaluation of Text-to-Audio-Video generation,
      covering visual/audio quality, synchronization, fine-grained controllability,
      physical plausibility, and holistic semantic alignment.
    </p>
  </div>
  <div class="header-links">
    <a href="https://github.com/microsoft/AVGen-Bench" target="_blank" rel="noopener">GitHub</a>
    <a href="https://arxiv.org/abs/2604.08540" target="_blank" rel="noopener">Paper</a>
    <a href="https://huggingface.co/datasets/microsoft/AVGen-Bench" target="_blank" rel="noopener">Dataset</a>
  </div>
</div>
"""


with gr.Blocks(title="AVGen-Bench Leaderboard") as demo:
    gr.HTML(HEADER)

    with gr.Tab("Leaderboard"):
        with gr.Row():
            component_type = gr.Dropdown(
                choices=["All", "Proprietary", "Open-source", "Mixed"],
                value="All",
                label="Component type",
            )
            sort_by = gr.Dropdown(
                choices=SORT_CHOICES,
                value="Total",
                label="Sort item",
            )
            sort_order = gr.Radio(
                choices=["Descending", "Ascending", "Best first"],
                value="Descending",
                label="Sort order",
            )
            query = gr.Textbox(label="Search", placeholder="Model or component")

        summary = gr.HTML()
        table = gr.HTML()

    with gr.Tab("Model Profile"):
        model = gr.Dropdown(choices=model_choices(LEADERBOARD), value=model_choices(LEADERBOARD)[0], label="Model")
        profile = gr.HTML()

    with gr.Tab("Metric Scheme"):
        gr.HTML(render_methodology())
        if OVERVIEW_IMAGE.exists():
            gr.Image(
                value=str(OVERVIEW_IMAGE),
                label="AVGen-Bench evaluation suite",
                show_label=False,
                interactive=False,
                elem_classes=["overview-image"],
            )

    with gr.Tab("Submission"):
        gr.HTML(
            """
            <p class="submission-copy">
              Submit raw AVGen-Bench metrics for review. The app recomputes
              <strong>Total</strong> from the raw metrics and sends the entry to a
              pending-review backend. Accepted entries are still merged into the
              official leaderboard manually.
            </p>
            """
        )
        with gr.Row():
            submit_model = gr.Textbox(label="Model name", placeholder="Your Model")
            submit_component_type = gr.Dropdown(
                choices=["Proprietary", "Open-source", "Mixed"],
                value="Open-source",
                label="Component type",
            )
        submit_components = gr.Textbox(
            label="Components",
            placeholder="VideoModel (Open-source)|AudioModel (Open-source)",
        )
        with gr.Row():
            submit_contact = gr.Textbox(label="Public contact", placeholder="GitHub handle or email")
            submit_model_url = gr.Textbox(label="Model or paper URL", placeholder="https://...")
            submit_results_url = gr.Textbox(label="Evaluation artifact URL", placeholder="https://...")
        submit_notes = gr.Textbox(label="Notes", lines=3, placeholder="Optional evaluation details")

        with gr.Accordion("Raw metric scores", open=True):
            with gr.Row():
                submit_vis = gr.Number(label="Vis")
                submit_aud = gr.Number(label="Aud (PQ)")
                submit_av = gr.Number(label="AV")
                submit_lip = gr.Number(label="Lip")
            with gr.Row():
                submit_text = gr.Number(label="Text")
                submit_face = gr.Number(label="Face")
                submit_music = gr.Number(label="Music")
                submit_speech = gr.Number(label="Speech")
            with gr.Row():
                submit_lophy = gr.Number(label="Lo-Phy")
                submit_hiphy = gr.Number(label="Hi-Phy")
                submit_holistic = gr.Number(label="Holistic")

        submit_button = gr.Button("Submit for Review", variant="primary")
        submit_status = gr.HTML()
        submit_payload = gr.Code(label="Submission JSON", language="json")
        gr.File(value=str(SUBMISSION_TEMPLATE), label="CSV template", interactive=False)

    demo.load(
        fn=update_leaderboard,
        inputs=[component_type, query, sort_by, sort_order],
        outputs=[summary, table],
    )
    demo.load(fn=update_profile, inputs=[model], outputs=[profile])

    for control in [component_type, query, sort_by, sort_order]:
        control.change(
            fn=update_leaderboard,
            inputs=[component_type, query, sort_by, sort_order],
            outputs=[summary, table],
        )

    model.change(fn=update_profile, inputs=[model], outputs=[profile])
    submit_button.click(
        fn=submit_score,
        inputs=[
            submit_model,
            submit_components,
            submit_component_type,
            submit_contact,
            submit_model_url,
            submit_results_url,
            submit_notes,
            submit_vis,
            submit_aud,
            submit_av,
            submit_lip,
            submit_text,
            submit_face,
            submit_music,
            submit_speech,
            submit_lophy,
            submit_hiphy,
            submit_holistic,
        ],
        outputs=[submit_status, submit_payload],
        api_name="submit_score",
    )


if __name__ == "__main__":
    launch_kwargs = {
        "css": CSS,
        "server_name": os.environ.get("GRADIO_SERVER_NAME", "0.0.0.0"),
    }
    port = os.environ.get("PORT") or os.environ.get("GRADIO_SERVER_PORT")
    if port:
        launch_kwargs["server_port"] = int(port)
    demo.launch(**launch_kwargs)