Update app.py

app.py

@@ -6,15 +6,10 @@ import os
 from PIL import Image
 
 import deepguard  # noqa: F401  # registers ms_eff_gcvit_* into timm
-import numpy as np
 from explainability import (
     HiResCAMExplainer, GradCAMElementWiseExplainer, LayerCAMExplainer,
     EigenGradCAMExplainer, GradCAMPlusPlusExplainer, XGradCAMExplainer,
 )
-from explainability.metrics.road import (
-    ROADMostRelevantFirstAverage, ROADLeastRelevantFirstAverage,
-)
-from explainability.utils.model_targets import BinaryClassifierOutputSigmoidTarget
 
 # ── Registries ────────────────────────────────────────────────────────────────
 MODEL_NAME = {"fast": "ms_eff_gcvit_b0", "pro": "ms_eff_gcvit_b5"}
@@ -61,62 +56,6 @@ def _get_explainer(variant: str, dataset: str, method: str, branch: str):
     return _cache[key]
 
 
-# ── ROAD metric table (video-style dark table) ────────────────────────────────
-def _metric_bar(val: float, color: str) -> str:
-    w = int(max(0.0, min(1.0, val)) * 100)
-    return (
-        '<div style="background:#0b0f19;border-radius:4px;height:14px;'
-        'width:160px;overflow:hidden;">'
-        f'<div style="background:{color};width:{w}%;height:100%;border-radius:4px;"></div></div>'
-    )
-
-
-def _build_metric_table(morf: float, lerf: float) -> str:
-    rows = [
-        ("MORF", "Most Relevant First", morf, "#f87171", "↓ lower = more faithful"),
-        ("LERF", "Least Relevant First", lerf, "#4ade80", "↑ higher = more faithful"),
-    ]
-    body = ""
-    for short, full, val, color, interp in rows:
-        body += f"""
-        <tr style="background:#12161f; border-bottom:1px solid #2d3748;">
-          <td style="padding:10px 14px; color:#FFD21E; font-weight:700;">{short}</td>
-          <td style="padding:10px 14px; color:#9ca3af;">{full}</td>
-          <td style="padding:10px 14px;">{_metric_bar(val, color)}</td>
-          <td style="padding:10px 14px; color:{color}; font-family:monospace; font-weight:700;">{val:.4f}</td>
-          <td style="padding:10px 14px; color:#9ca3af; font-size:13px;">{interp}</td>
-        </tr>"""
-
-    return f"""
-    <div style="overflow-x:auto; margin-top:8px;">
-      <table style="width:100%; border-collapse:collapse; font-size:14px;
-                    background:#0b0f19; border-radius:8px; overflow:hidden;">
-        <thead>
-          <tr style="background:#FFD21E; color:#1a1a1a;">
-            <th style="padding:10px 14px; text-align:left;">Metric</th>
-            <th style="padding:10px 14px; text-align:left;">Perturbation</th>
-            <th style="padding:10px 14px; text-align:left;">Faithfulness</th>
-            <th style="padding:10px 14px; text-align:left;">Score</th>
-            <th style="padding:10px 14px; text-align:left;">Interpretation</th>
-          </tr>
-        </thead>
-        <tbody>{body}</tbody>
-      </table>
-    </div>"""
-
-
-def _evaluate_road(explainer, img_path: str) -> str:
-    torch.manual_seed(2026)
-    np.random.seed(2026)
-    pct = [20, 40, 60, 80]  # reduced from 9 → 4 for speed
-    grayscale_cam, tensor, _ = explainer._build_grayscale_cam(img_path)
-    cams = grayscale_cam[np.newaxis, ...]        # (1, H, W)
-    targets = [BinaryClassifierOutputSigmoidTarget()]
-    morf = float(ROADMostRelevantFirstAverage(pct)(tensor, cams, targets, explainer.model)[0])
-    lerf = float(ROADLeastRelevantFirstAverage(pct)(tensor, cams, targets, explainer.model)[0])
-    return _build_metric_table(morf, lerf)
-
-
 # ── Inference ─────────────────────────────────────────────────────────────────
 @spaces.GPU
 def explain(
@@ -170,31 +109,6 @@ def explain(
     return result
 
 
-# ── ROAD evaluation (separate button) ─────────────────────────────────────────
-@spaces.GPU
-def evaluate_metrics(image, variant, dataset, branch, method):
-    if image is None:
-        raise gr.Error("Please upload an image first.")
-
-    with tempfile.NamedTemporaryFile(suffix=".jpg", delete=False) as f:
-        image.save(f.name)
-        tmp_path = f.name
-
-    try:
-        explainer = _get_explainer(variant, dataset, method, branch)
-        if torch.cuda.is_available():
-            explainer.model.to("cuda:0")
-            explainer.device = "cuda:0"
-        return _evaluate_road(explainer, tmp_path)
-    except Exception as e:
-        msg = str(e)
-        if "face" in msg.lower() or "NoneType" in msg or "index" in msg.lower():
-            raise gr.Error("😶 No face detected. Please use an image with a clearly visible face.")
-        raise gr.Error(f"ROAD evaluation error: {e}")
-    finally:
-        os.unlink(tmp_path)
-
-
 # ── Assets (HF CDN absolute URL — keeps GIF animation) ─────────────────────────
 _HF_BASE = "https://huggingface.co/spaces/KoreaPeter/DeepFake-Detection-XAI/resolve/main"
 DUAL_BRANCH_SRC = f"{_HF_BASE}/dual_branch.gif"
@@ -410,15 +324,6 @@ with gr.Blocks(title="DeepGuard XAI") as demo:
                 info="PCA noise reduction → keeps dominant forgery pattern only.",
             )
 
-    with gr.Accordion("📊 Explainability Metrics — ROAD", open=True):
-        gr.Markdown(
-            "CAM **faithfulness** via ROAD (perturbation-based). "
-            "Pixels are removed by relevance and the drop in prediction confidence is measured. "
-            "⚠️ Heavier than visualization — run only when needed."
-        )
-        eval_btn = gr.Button("📊 Evaluate ROAD", variant="secondary")
-        metric_output = gr.HTML("")
-
     # branch → method choices sync (recommended method becomes default)
     def _sync_methods(branch):
         choices = METHODS_BY_BRANCH[branch]
@@ -446,11 +351,6 @@ with gr.Blocks(title="DeepGuard XAI") as demo:
         image_weight_slider, threshold_slider,
     ]
     run_btn.click(fn=explain, inputs=inputs, outputs=result_output)
-    eval_btn.click(
-        fn=evaluate_metrics,
-        inputs=[image_input, variant_radio, dataset_radio, branch_radio, method_radio],
-        outputs=metric_output,
-    )
 
 if __name__ == "__main__":
     demo.launch(theme=THEME, css=CSS)