Upload app.py

2.CNN/app.py

@@ -35,6 +35,9 @@ forward_prop = training_mod.forward_prop
 get_predictions = training_mod.get_predictions
 softmax = training_mod.softmax
 
+# Detect if running on Hugging Face Spaces
+IS_SPACE = bool(os.getenv("SPACE_ID"))
+
 
 def _metric_status(name, value):
     target = METRIC_TARGETS.get(name)
@@ -128,36 +131,43 @@ def shift_with_zero_pad(arr, shift_y=0, shift_x=0):
 
 
 def dilate_binary_like(arr, radius=1):
-    pad = radius
-    padded = np.pad(arr, pad, mode="constant", constant_values=0.0)
-    out = np.zeros_like(arr)
-    for i in range(arr.shape[0]):
-        for j in range(arr.shape[1]):
-            window = padded[i : i + 2 * pad + 1, j : j + 2 * pad + 1]
-            out[i, j] = window.max()
-    return out
+    # Vectorized dilation via max over shifted windows (3x3 when radius=1)
+    if radius != 1:
+        # Fallback to radius=1 behavior for simplicity/perf
+        radius = 1
+    shifts = []
+    for dy in (-1, 0, 1):
+        for dx in (-1, 0, 1):
+            shifts.append(shift_with_zero_pad(arr, dy, dx))
+    stacked = np.stack(shifts, axis=0)
+    return np.max(stacked, axis=0)
 
 
 def erode_binary_like(arr, radius=1):
-    pad = radius
-    padded = np.pad(arr, pad, mode="constant", constant_values=1.0)
-    out = np.zeros_like(arr)
-    for i in range(arr.shape[0]):
-        for j in range(arr.shape[1]):
-            window = padded[i : i + 2 * pad + 1, j : j + 2 * pad + 1]
-            out[i, j] = window.min()
-    return out
+    # Vectorized erosion via min over shifted windows (3x3 when radius=1)
+    if radius != 1:
+        radius = 1
+    shifts = []
+    for dy in (-1, 0, 1):
+        for dx in (-1, 0, 1):
+            shifts.append(shift_with_zero_pad(arr, dy, dx))
+    stacked = np.stack(shifts, axis=0)
+    return np.min(stacked, axis=0)
 
 
-def generate_inference_variants(arr):
+def generate_inference_variants(arr, *, fast: bool = False):
     variants = []
-    # slight shifts
+    if fast:
+        # Space-optimized: only cardinal shifts (4 variants)
+        for dy, dx in ((-1, 0), (1, 0), (0, -1), (0, 1)):
+            variants.append(shift_with_zero_pad(arr, dy, dx))
+        return variants
+    # Full set: 8 shifts + morphology
     for dy in (-1, 0, 1):
         for dx in (-1, 0, 1):
             if dy == 0 and dx == 0:
                 continue
             variants.append(shift_with_zero_pad(arr, dy, dx))
-    # dilation/erosion to handle thin or thick strokes
     variants.append(dilate_binary_like(arr, radius=1))
     variants.append(erode_binary_like(arr, radius=1))
     return variants
@@ -298,7 +308,7 @@ def preprocess_image(img_input, stroke_scale: float = 1.0, *, auto_balance: bool
             clamp=(0.7, 1.4),
         )
 
-    augmented_arrays = [arr_resized, *generate_inference_variants(arr_resized)]
+    augmented_arrays = [arr_resized, *generate_inference_variants(arr_resized, fast=IS_SPACE)]
     augmented_standardized = [
         (arr.reshape(TARGET_HEIGHT * TARGET_WIDTH, 1) - mean) / std_safe
         for arr in augmented_arrays
@@ -534,7 +544,7 @@ def predict_number(left_canvas, right_canvas, stroke_scale, auto_balance):
     return pred, prob_rows, (preview * 255).astype(np.uint8), mean_diff, json.dumps(diagnostics, indent=2)
 
 
-with gr.Blocks() as demo:
+with gr.Blocks(queue=True) as demo:
     gr.Markdown(
         """
         # Elliot's MNIST-100 Classifier
@@ -591,24 +601,26 @@ with gr.Blocks() as demo:
         inputs=[left_canvas, right_canvas, stroke_slider, auto_balance],
         outputs=[pred_box, prob_table, preview, mean_diff_view, diagnostics_box],
     )
-    left_canvas.change(
-        fn=predict_number,
-        inputs=[left_canvas, right_canvas, stroke_slider, auto_balance],
-        outputs=[pred_box, prob_table, preview, mean_diff_view, diagnostics_box],
-    )
-    right_canvas.change(
-        fn=predict_number,
-        inputs=[left_canvas, right_canvas, stroke_slider, auto_balance],
-        outputs=[pred_box, prob_table, preview, mean_diff_view, diagnostics_box],
-    )
+    # On Spaces, avoid per-stroke inference to prevent event floods
+    if not IS_SPACE:
+        left_canvas.change(
+            fn=predict_number,
+            inputs=[left_canvas, right_canvas, stroke_slider, auto_balance],
+            outputs=[pred_box, prob_table, preview, mean_diff_view, diagnostics_box],
+        )
+        right_canvas.change(
+            fn=predict_number,
+            inputs=[left_canvas, right_canvas, stroke_slider, auto_balance],
+            outputs=[pred_box, prob_table, preview, mean_diff_view, diagnostics_box],
+        )
 
 
 if __name__ == "__main__":
     space_env = os.getenv("SPACE_ID")
     if space_env:
-        demo.launch(show_api=False)
+        demo.queue(concurrency_count=1).launch(show_api=False)
     else:
-        demo.launch(server_name="0.0.0.0", share=True, show_api=False)
+        demo.queue(concurrency_count=1).launch(server_name="0.0.0.0", share=True, show_api=False)
 def _disable_gradio_api_schema(*_args, **_kwargs):
     """Work around Gradio schema bug on Python 3.13 by returning empty metadata."""
     return {}