move api helpers to api_helper.py

api_helper.py

@@ -0,0 +1,56 @@
+from PIL import Image
+import numpy as np
+import base64
+import json
+from torchvision.transforms import Compose, Resize, CenterCrop
+
+# support sending images as base64
+
+def encode_numpy_array(image_np):
+    # Flatten the numpy array and convert it to bytes
+    image_bytes = image_np.tobytes()
+
+    # Encode the byte data as base64
+    encoded_image = base64.b64encode(image_bytes).decode()
+    payload = {
+        "encoded_image": encoded_image,
+        "width": image_np.shape[1],
+        "height": image_np.shape[0],
+        "channels": image_np.shape[2],
+    }
+    payload_json = json.dumps(payload)
+    return payload_json
+
+def decode_numpy_array(payload):
+    payload_json = json.loads(payload)
+    # payload_json = payload.json()
+    encoded_image = payload_json["encoded_image"]
+    width = payload_json["width"]
+    height = payload_json["height"]
+    channels = payload_json["channels"]
+    # Decode the base64 data
+    decoded_image = base64.b64decode(encoded_image)
+
+    # Convert the byte data back to a NumPy array
+    image_np = np.frombuffer(decoded_image, dtype=np.uint8).reshape(height, width, channels)
+
+    return image_np
+
+
+def preprocess_image(image_np, max_size=224):
+    # Convert the numpy array to a PIL image
+    image = Image.fromarray(image_np)
+
+    # Define the transformation pipeline
+    transforms = Compose([
+        Resize(max_size, interpolation=Image.BICUBIC),
+        CenterCrop(max_size),
+    ])
+
+    # Apply the transformations to the image
+    image = transforms(image)
+
+    # Convert the PIL image back to a numpy array
+    image_np = np.array(image)
+
+    return image_np

app.py

@@ -9,6 +9,8 @@ import math
 # from transformers import CLIPTextModel, CLIPTokenizer
 import os
 
+from api_helper import encode_numpy_array, decode_numpy_array, preprocess_image
+
 # clip_model_id = "openai/clip-vit-large-patch14-336"
 # clip_retrieval_indice_name, clip_model_id ="laion5B-L-14", "/laion/CLIP-ViT-L-14-laion2B-s32B-b82K"
 clip_retrieval_service_url = "https://knn.laion.ai/knn-service"
@@ -35,61 +37,7 @@ def debug_print(*args, **kwargs):
     if debug_print_on:
         print(*args, **kwargs)
 
-# support sending images as base64
-
-def encode_numpy_array(image_np):
-    import base64
-    import json
-    # Flatten the numpy array and convert it to bytes
-    image_bytes = image_np.tobytes()
-
-    # Encode the byte data as base64
-    encoded_image = base64.b64encode(image_bytes).decode()
-    payload = {
-        "encoded_image": encoded_image,
-        "width": image_np.shape[1],
-        "height": image_np.shape[0],
-        "channels": image_np.shape[2],
-    }
-    payload_json = json.dumps(payload)
-    return payload_json
-
-def decode_numpy_array(payload):
-    import base64
-    import json
-    payload_json = json.loads(payload)
-    # payload_json = payload.json()
-    encoded_image = payload_json["encoded_image"]
-    width = payload_json["width"]
-    height = payload_json["height"]
-    channels = payload_json["channels"]
-    # Decode the base64 data
-    decoded_image = base64.b64decode(encoded_image)
-
-    # Convert the byte data back to a NumPy array
-    image_np = np.frombuffer(decoded_image, dtype=np.uint8).reshape(height, width, channels)
-
-    return image_np
-
-
-def preprocess_image(image_np, max_size=224):
-    from torchvision.transforms import Compose, Resize, CenterCrop
-    # Convert the numpy array to a PIL image
-    image = Image.fromarray(image_np)
-
-    # Define the transformation pipeline
-    transforms = Compose([
-        Resize(max_size, interpolation=Image.BICUBIC),
-        CenterCrop(max_size),
-    ])
-
-    # Apply the transformations to the image
-    image = transforms(image)
-
-    # Convert the PIL image back to a numpy array
-    image_np = np.array(image)
 
-    return image_np
 
 def image_to_embedding(input_im):
     # debug_print("image_to_embedding")