Update app.py

app.py

@@ -465,7 +465,7 @@ def classify_crops(
 
 
 # -----------------------------------------------------------------------------
-# mT5 source serialization (aligned with train_boxes_seq2seq_pipeline.py)
+# mT5 source serialization
 # -----------------------------------------------------------------------------
 
 
@@ -894,18 +894,18 @@ with gr.Blocks(title="Paleo‑Hebrew Tablet Reader", css=CSS) as demo:
             inp = gr.Image(type="pil", label="Input Image", height=IMG_HEIGHT)
 
             with gr.Accordion("Vision Settings", open=True):
-                conf = gr.Slider(0.05, 0.95, value=0.25, step=0.01, label="YOLO conf")
-                iou = gr.Slider(0.10, 0.90, value=0.45, step=0.01, label="YOLO IoU")
+                conf = gr.Slider(0.05, 0.95, value=0.25, step=0.01, label="Box Confidence")
+                iou = gr.Slider(0.10, 0.90, value=0.45, step=0.01, label="IoU")
                 max_det = gr.Slider(1, 200, value=80, step=1, label="Max Detections")
                 crop_pad = gr.Slider(0.0, 0.8, value=0.20, step=0.01, label="Crop Pad")
                 topk_k = gr.Slider(1, 10, value=5, step=1, label="Classifier Top‑K")
 
             with gr.Accordion("mT5 + Translation Settings", open=True):
-                rtl = gr.Checkbox(value=True, label="RTL order (right→left) [for reading-order mode]")
+                rtl = gr.Checkbox(value=True, label="RTL order (right to left) [for reading-order mode]")
                 order_mode = gr.Radio(["reading", "detector"], value="reading", label="Box order")
                 coord_norm = gr.Radio(["boxes", "det", "none"], value="boxes", label="coord_norm (like training)")
                 output_mode = gr.Radio(["he", "en_direct", "he_then_en"], value="he_then_en", label="Output mode")
-                he2en_kind = gr.Dropdown(choices=TRANSLATOR_CHOICES, value="opus", label="Hebrew → English translator")
+                he2en_kind = gr.Dropdown(choices=TRANSLATOR_CHOICES, value="opus", label="Hebrew to English translator")
 
             with gr.Accordion("How to run a VLM yourself (optional)", open=False):
                 gr.Markdown(
@@ -949,12 +949,12 @@ pred = proc.batch_decode(out, skip_special_tokens=True)[0]
                         out_dbg_pipe = gr.Code(label="Debug JSON", language="json")
                         out_mt5_src = gr.Textbox(label="mT5 source text (fed into mT5)", lines=10, interactive=False)
 
-                with gr.Tab("Detector (YOLO)"):
+                with gr.Tab("Detector"):
                     run_det_btn = gr.Button("Run Detector", variant="primary")
                     out_fig_det = gr.Plot(label="Detected Bounding Boxes")
                     out_json_det = gr.Code(label="Raw Detection Output", language="json")
 
-                with gr.Tab("Classifier (ConvNeXt)"):
+                with gr.Tab("Classifier"):
                     run_cls_btn = gr.Button("Run Classifier", variant="primary")
                     out_crops_cls = gr.Gallery(columns=8, height=360, label="Isolated Letter Crops")
                     out_json_cls = gr.Code(label="Top‑K per Box", language="json")