Add Gradio app, requirements and model

.gitattributes

@@ -33,3 +33,4 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *.zip filter=lfs diff=lfs merge=lfs -text
 *.zst filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
+cxp_projection_rotation.pt filter=lfs diff=lfs merge=lfs -text

app.py

@@ -0,0 +1,100 @@
+import os
+import numpy as np
+import onnxruntime as ort
+import gradio as gr
+from PIL import Image
+
+# モデルをONNX Runtimeで読み込む
+session = ort.InferenceSession("cxp_projection_rotation.pt", None)
+input_name = session.get_inputs()[0].name
+# 出力ヘッド名（順序に注意）
+projection_name = session.get_outputs()[0].name
+rotation_name = session.get_outputs()[1].name
+
+# クラスラベルの定義
+PROJ_LABELS = ["AP", "PA", "Lateral"]
+ROT_LABELS  = ["Upright", "Inverted", "Left90", "Right90"]
+
+def softmax(x: np.ndarray) -> np.ndarray:
+    """数値安定版ソフトマックス"""
+    e = np.exp(x - np.max(x))
+    return e / np.sum(e)
+
+def predict(image_path: str) -> str:
+    # 画像をグレースケールで読み込み（Lモード）
+    img = Image.open(image_path).convert("L")
+    arr = np.array(img, dtype=np.float32) / 255.0        # 正規化
+    arr = arr[np.newaxis, np.newaxis, :, :]              # [1,1,H,W] バッチ＆チャネル次元追加
+
+    # 推論
+    proj_logits, rot_logits = session.run(
+        [projection_name, rotation_name],
+        {input_name: arr}
+    )
+    proj_probs = softmax(proj_logits[0])
+    rot_probs  = softmax(rot_logits[0])
+
+    # 予測結果と確率
+    proj_idx  = int(np.argmax(proj_probs))
+    rot_idx   = int(np.argmax(rot_probs))
+    proj_lbl  = PROJ_LABELS[proj_idx]
+    rot_lbl   = ROT_LABELS[rot_idx]
+    proj_p    = proj_probs[proj_idx]
+    rot_p     = rot_probs[rot_idx]
+
+    # ファイル名から元ラベルを推定（例: "AP_Upright.png" → ["AP","Upright"]）
+    base = os.path.splitext(os.path.basename(image_path))[0]
+    try:
+        orig_proj, orig_rot = base.split("_", 1)
+    except ValueError:
+        orig_proj = orig_rot = None
+
+    # 警告メッセージ（必要なら赤字で表示）
+    warnings_html = ""
+    if orig_proj and orig_proj != proj_lbl:
+        warnings_html += "<p style='color:red'>⚠ Potential mislabeled projection</p>"
+    if orig_rot and orig_rot != rot_lbl:
+        warnings_html += "<p style='color:red'>⚠ Potential mislabeled rotation</p>"
+
+    # HTML形式で結果を返す
+    html = (
+        f"<p><strong>Projection :</strong> {proj_lbl} (p={proj_p:.3f})</p>"
+        f"<p><strong>Rotation   :</strong> {rot_lbl} (p={rot_p:.3f})</p>"
+        f"{warnings_html}"
+    )
+    return html
+
+# Gradio UI構築
+with gr.Blocks() as demo:
+    with gr.Row():
+        with gr.Column():
+            # 画像アップロード（PNG/L, 256×256前提）
+            image_input = gr.Image(
+                label="Upload PNG (256×256)", 
+                type="filepath", 
+                tool=None
+            )
+            # sample_imagesフォルダから4枚まで例示
+            sample_list = sorted(
+                [os.path.join("sample_images", f) 
+                 for f in os.listdir("sample_images") 
+                 if f.lower().endswith(".png")]
+            )
+            gr.Examples(
+                examples=sample_list, 
+                inputs=image_input, 
+                label="Sample Images"
+            )
+        with gr.Column():
+            # 推論結果用HTML表示エリア
+            result = gr.HTML()
+
+    # 画像選択・アップロード時に自動でpredictを実行
+    image_input.change(
+        fn=predict, 
+        inputs=image_input, 
+        outputs=result
+    )
+
+if __name__ == "__main__":
+    demo.launch()

cxp_projection_rotation.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:67d5e766adf1acec37c112e87d568ffda2c8a1fe9334f1c0ab3ef3dbfe35aae4
+size 71017114

requirements.txt

@@ -0,0 +1,4 @@
+gradio
+onnxruntime
+numpy
+Pillow