Create app.py

app.py

@@ -0,0 +1,114 @@
+import gradio as gr
+from models import text_model, TEXT_FRIENDLY_MAP, image_model
+
+# -----------------------------
+# Helpers
+# -----------------------------
+def _friendly_text_label(raw_label: str) -> str:
+    # Normalize raw label from the text model to our friendly label
+    raw = (raw_label or "").strip().upper()
+    return TEXT_FRIENDLY_MAP.get(raw_label, TEXT_FRIENDLY_MAP.get(raw, raw_label))
+
+def _to_label_dict(items, friendly=False, is_text=False):
+    """
+    Convert pipeline outputs (list of dicts) to a {label: score} dict for gr.Label.
+
+    items: e.g. [{"label": "POSITIVE", "score": 0.98}, ...]
+    friendly: if True and is_text=True, map to friendly 'Real / Credible' or 'Fake / Not Credible'
+    """
+    out = {}
+    for it in items:
+        lab = it.get("label", "")
+        if friendly and is_text:
+            lab = _friendly_text_label(lab)
+        out[lab] = float(it.get("score", 0.0))
+    return out
+
+# -----------------------------
+# Inference functions
+# -----------------------------
+def predict_text(text: str):
+    if not text or not text.strip():
+        return "Please paste some text first.", {"Waiting...": 1.0}
+
+    # Get top-2 to show a nice confidence breakdown
+    preds = text_model(text, top_k=2)
+    # When top_k is used, transformers returns a list of dicts directly
+    # Normalize to dict for gr.Label component
+    label_dict = _to_label_dict(preds, friendly=True, is_text=True)
+
+    # Pick top result (highest score)
+    top_label, top_score = max(label_dict.items(), key=lambda kv: kv[1])
+    summary = f"{top_label} ({round(top_score * 100, 2)}%)"
+    return summary, label_dict
+
+def predict_image(image):
+    if image is None:
+        return "Please upload an image first.", {"Waiting...": 1.0}
+
+    # Get top-5 predictions for a nicer breakdown
+    preds = image_model(image, top_k=5)
+    label_dict = _to_label_dict(preds, friendly=False, is_text=False)
+
+    top_label, top_score = max(label_dict.items(), key=lambda kv: kv[1])
+    summary = f"{top_label} ({round(top_score * 100, 2)}%)"
+    return summary, label_dict
+
+# -----------------------------
+# UI
+# -----------------------------
+with gr.Blocks(title="AI Detection Hub") as demo:
+    gr.Markdown(
+        """
+# 🔎 AI Detection Hub
+Analyze **text** (credibility demo) and **images** (generic classifier) with a clean, simple interface.
+
+> ✅ Uses models that are known to load on Spaces without errors.  
+> 🧩 Later, you can swap in your own fine-tuned fake-news and deepfake detectors.
+        """
+    )
+
+    with gr.Tab("📝 Text Analysis"):
+        gr.Markdown("Paste a claim or snippet. The model returns a **credibility-style** score (demo using sentiment).")
+        with gr.Row():
+            text_input = gr.Textbox(
+                label="Enter text",
+                placeholder="Paste news text or a claim here...",
+                lines=6
+            )
+        with gr.Row():
+            text_btn = gr.Button("Analyze", variant="primary")
+            text_clear = gr.Button("Clear")
+        with gr.Row():
+            text_summary = gr.Textbox(label="Result", interactive=False)
+            text_probs = gr.Label(label="Confidence")
+
+        examples = gr.Examples(
+            label="Try examples",
+            examples=[
+                ["Breaking: Local council approves new park renovation after community vote."],
+                ["Celebrity endorses miracle cure that doctors say is unproven and dangerous."],
+            ],
+            inputs=[text_input],
+        )
+
+        text_btn.click(predict_text, inputs=text_input, outputs=[text_summary, text_probs])
+        text_clear.click(lambda: ("", "", {}), outputs=[text_input, text_summary, text_probs])
+
+    with gr.Tab("🖼️ Image Analysis"):
+        gr.Markdown("Upload an image. The model returns **top predicted classes** (generic classifier demo).")
+        with gr.Row():
+            image_input = gr.Image(type="pil", label="Upload image")
+        with gr.Row():
+            image_btn = gr.Button("Analyze", variant="primary")
+            image_clear = gr.Button("Clear")
+        with gr.Row():
+            image_summary = gr.Textbox(label="Result", interactive=False)
+            image_probs = gr.Label(label="Top-5 Confidence")
+
+        image_btn.click(predict_image, inputs=image_input, outputs=[image_summary, image_probs])
+        image_clear.click(lambda: (None, "", {}), outputs=[image_input, image_summary, image_probs])
+
+# Required entry point
+if __name__ == "__main__":
+    demo.launch()