Update app.py

app.py

@@ -479,21 +479,21 @@ def run(orig_image: Image.Image, points: list, preset_name: str):
 
     if orig_image is None:
         gr.Warning("Upload an image first.")
-        return None, "", None, None
+        return None, "", None
     if not points or len(points) < 2:
         gr.Warning("Provide at least 2 fixations (click the image or load a fixation file).")
-        return None, "", None, None
+        return None, "", None
 
     preset_key = PRESETS[preset_name]
     try:
         predictor = get_predictor(preset_key)
     except FileNotFoundError as e:
         gr.Warning(str(e))
-        return None, f"**Checkpoint not found** for preset `{preset_key}`.", None, None
+        return None, f"**Checkpoint not found** for preset `{preset_key}`.", None
     except Exception as e:  # noqa: BLE001
         traceback.print_exc()
         gr.Warning(f"Could not load model: {e}")
-        return None, "", None, None
+        return None, "", None
 
     w, h = orig_image.size
     # Build a MIMIC-style scanpath dataframe. The 3rd component (weight)
@@ -512,7 +512,7 @@ def run(orig_image: Image.Image, points: list, preset_name: str):
     scanpath = get_scanpath(df, "webdemo", img_height=h, img_width=w)
     if scanpath is None or scanpath.numel() == 0:
         gr.Warning("Could not build a scanpath from the fixations.")
-        return None, "", None, None
+        return None, "", None
     scanpath = scanpath[:200].to(predictor.device)
 
     img_tensor = predictor.transform(np.array(orig_image)).unsqueeze(0).to(predictor.device)
@@ -529,7 +529,7 @@ def run(orig_image: Image.Image, points: list, preset_name: str):
     except Exception as e:  # noqa: BLE001
         traceback.print_exc()
         gr.Warning(f"Prediction failed: {e}")
-        return None, "", None, None
+        return None, "", None
 
     class_probs = {c: float(p) for c, p in zip(predictor.classes, probs)}
     predicted_class = max(class_probs, key=class_probs.get)
@@ -540,14 +540,13 @@ def run(orig_image: Image.Image, points: list, preset_name: str):
     display_img = np.array(orig_image.resize((img_size, img_size)))
     gaze_hm = gaze_duration_heatmap(points, img_size, img_size, w, h)
     overlay = make_overlay(display_img, gaze_hm)
-    heatmap_img = heatmap_to_image(gaze_hm)
 
     prob_lines = "\n".join(
         f"- **{c}**: {p:.3f}" for c, p in sorted(class_probs.items(), key=lambda kv: -kv[1])
     )
     summary = f"### Predicted: **{predicted_class}**\n\n{prob_lines}"
 
-    return class_probs, summary, overlay, heatmap_img
+    return class_probs, summary, overlay
 
 
 # ─────────────────────────────────────────────────────────────────────────────
@@ -607,24 +606,25 @@ with gr.Blocks(title="GazeAlign", css=_CSS) as demo:
                 clear_btn = gr.Button("Clear fixations")
                 delete_btn = gr.Button("Delete image", visible=False)
 
-            gr.Markdown("**— or —** load fixations from a file:")
-            fixfile = gr.File(
-                label=_FIXFILE_LABEL, file_types=[".csv", ".xlsx", ".xls"], type="filepath"
-            )
-            with gr.Row(visible=False) as mapping_row:
-                id_col_dd = gr.Dropdown(label="ID column", choices=[])
-                x_col_dd = gr.Dropdown(label="X column", choices=[])
-                y_col_dd = gr.Dropdown(label="Y column", choices=[])
-                time_col_dd = gr.Dropdown(label="Time column (optional)", choices=[])
-            apply_fix_btn = gr.Button("Apply fixation file", visible=False)
+            # Collapsed by default — click the header (menu-style) to reveal.
+            with gr.Accordion("📁  Load fixations from a file (.csv / .xlsx / .xls)", open=False):
+                fixfile = gr.File(
+                    label=_FIXFILE_LABEL, file_types=[".csv", ".xlsx", ".xls"], type="filepath"
+                )
+                with gr.Row(visible=False) as mapping_row:
+                    id_col_dd = gr.Dropdown(label="ID column", choices=[])
+                    x_col_dd = gr.Dropdown(label="X column", choices=[])
+                    y_col_dd = gr.Dropdown(label="Y column", choices=[])
+                    time_col_dd = gr.Dropdown(label="Time column (optional)", choices=[])
+                apply_fix_btn = gr.Button("Apply fixation file", visible=False)
+
+            run_btn = gr.Button("Run GazeAlign", variant="primary", elem_id="run-btn")
 
         # ── Right: results ───────────────────────────────────────────────────
         with gr.Column(scale=1):
-            run_btn = gr.Button("Run GazeAlign", variant="primary", elem_id="run-btn")
             label_output = gr.Label(label="Predicted class (probabilities)", num_top_classes=5)
             summary_output = gr.Markdown()
             overlay_output = gr.Image(label="Gaze-fixation heatmap (dwell-weighted) — overlay")
-            mask_output = gr.Image(label="Gaze-fixation heatmap (weighted by duration)")
 
     gr.Markdown(
         "<div class='footer-note'>Fixation tables use raw pixel or normalised "
@@ -665,7 +665,7 @@ with gr.Blocks(title="GazeAlign", css=_CSS) as demo:
     run_btn.click(
         run,
         [orig_image_state, points_state, preset_dd],
-        [label_output, summary_output, overlay_output, mask_output],
+        [label_output, summary_output, overlay_output],
     )