Update app.py

app.py

@@ -2,54 +2,62 @@ import gradio as gr
 from roboflow import Roboflow
 import tempfile
 import os
-import cv2
-import supervision as sv
-import numpy as np
+from sahi.slicing import slice_image
+from sahi import get_sliced_prediction
+from sahi.postprocess import postprocess_predictions
 
-# Inisialisasi Roboflow
+# Inisialisasi Roboflow (for model path)
 rf = Roboflow(api_key="Otg64Ra6wNOgDyjuhMYU")
 project = rf.workspace("alat-pelindung-diri").project("nescafe-4base")
 model = project.version(16).model
 
-# Fungsi untuk deteksi objek dengan supervision InferenceSlicer
+# Fungsi untuk deteksi objek menggunakan SAHI
 def detect_objects(image):
-    # Simpan gambar sementara
+    # Menyimpan gambar sementara
     with tempfile.NamedTemporaryFile(delete=False, suffix=".jpg") as temp_file:
         image.save(temp_file, format="JPEG")
         temp_file_path = temp_file.name
 
-    # Membaca gambar dengan OpenCV
-    img = cv2.imread(temp_file_path)
+    # Slice gambar menjadi potongan-potongan kecil
+    slice_image_result = slice_image(
+        image=temp_file_path,
+        output_file_name="sliced_image",
+        output_dir="/tmp/sliced/",
+        slice_height=256,
+        slice_width=256,
+        overlap_height_ratio=0.1,
+        overlap_width_ratio=0.1
+    )
 
-    # Callback function untuk model prediksi
-    def callback(image_slice: np.ndarray) -> sv.Detections:
-        # Lakukan inferensi pada setiap potongan gambar
-        predictions = model.predict(image_slice, confidence=50, overlap=30).json()
-        return sv.Detections.from_inference(predictions)
+    # Jalankan prediksi pada setiap potongan gambar
+    predictions = get_sliced_prediction(
+        image_path=temp_file_path,
+        model_path="path_to_your_model",
+        model_type="yolov5",  # Adjust based on your model
+        slice_height=256,
+        slice_width=256,
+        overlap_height_ratio=0.1,
+        overlap_width_ratio=0.1,
+        model_confidence_threshold=0.25
+    )
 
-    # Menggunakan InferenceSlicer
-    slicer = sv.InferenceSlicer(callback=callback, overlap_wh=(0.2, 0.2))  # Replacing overlap_ratio_wh
+    # Postprocess dan gabungkan hasil prediksi
+    postprocessed_predictions = postprocess_predictions(
+        predictions=predictions,
+        postprocess_type='NMS',
+        iou_threshold=0.5
+    )
 
-    # Proses gambar dengan slicer
-    detections = slicer(img)
+    # Annotate gambar dengan hasil prediksi
+    annotated_image = model.annotate_image_with_predictions(temp_file_path, postprocessed_predictions)
 
-    # Filter deteksi yang tumpang tindih (gunakan NMM atau tanpa filter)
-    overlap_filter = sv.OverlapFilter(strategy=sv.OverlapFilter.NON_MAX_MERGE, iou_threshold=0.5)
-    filtered_detections = overlap_filter.filter(detections)
-
-    # Annotasi gambar dengan deteksi
-    annotated_image = sv.BoxAnnotator().annotate(scene=img.copy(), detections=filtered_detections)
-
-    # Simpan gambar dengan prediksi
-    output_path = "/tmp/prediction.jpg"
-    cv2.imwrite(output_path, annotated_image)
-
-    # Hapus file sementara
-    os.remove(temp_file_path)
+    # Simpan gambar hasil annotasi
+    output_image_path = "/tmp/prediction.jpg"
+    annotated_image.save(output_image_path)
 
     # Menghitung jumlah objek per kelas
     class_count = {}
-    for detection in filtered_detections:
+    for detection in postprocessed_predictions:
         class_name = detection.class_name
         if class_name in class_count:
             class_count[class_name] += 1
@@ -61,7 +69,10 @@ def detect_objects(image):
     for class_name, count in class_count.items():
         result_text += f"{class_name}: {count} objek\n"
 
-    return output_path, result_text
+    # Hapus file sementara
+    os.remove(temp_file_path)
+
+    return output_image_path, result_text
 
 # Membuat antarmuka Gradio
 iface = gr.Interface(