0

我有一个非常简单的 CNN 模型，一个多标签图像分类器，我在 Colab 笔记本中进行了预训练。我保存了模型和权重，并将其作为模型上传到 Google Cloud AI Platform（统一）中。

我正在使用 Cloud Functions 从 URL 下载图像并使用 Google 的aiplatformPython 库将其传递给分类器。

分类器的输入层采用[1, 180, 180, 3]（具有 3 个通道的 180x180 图像）的形状。

我用来将其发送到我的模型的代码基于Google 的官方示例。

我的问题是：instances下面的代码需要是什么样的？我是否需要以某种方式将其解析为 JSON 或其他格式？我只是不明白这个对象应该是什么才能把它放到我的模型中。

此代码使用 Google 的aiplatformPython 库。

# Code used to download and transform the image to a shape of [1, 180, 180, 3]
def load_image(url):
    response = requests.get(url)
    img_bytes = BytesIO(response.content)
    img = Image.open(img_bytes)
    img = img.convert('RGB')
    img = img.resize((180,180), Image.NEAREST)
    arr = preprocessing.image.img_to_array(img)
    arr = tf.expand_dims(arr, 0)
    return arr

发送预测：

if request_json['url']:
   url = request_json['url']
   image_array = load_image(url)
        
   client_options = {"api_endpoint": API_ENDPOINT}
   client = aiplatform.gapic.PredictionServiceClient(client_options=client_options)

我不明白示例中接下来的两行中发生了什么。 我的模型需要了解 image_bytes 吗？我应该向它发送base64吗？

   instance_dict = {"image_bytes": {"b64": image_array}, "key": "0"}     
   instance = json_format.ParseDict(instance_dict, Value())
   
   instances=[image_array]
   parameters_dict = {}
   parameters = json_format.ParseDict(parameters_dict, Value())
   endpoint = client.endpoint_path(
      project=PROJECT_ID, location=LOCATION, endpoint=ENDPOINT_ID
   ) 
        
   response = client.predict(
      endpoint=endpoint, instances=instances, parameters=parameters
   )
   
   print("response")
   print(" deployed_model_id:", response.deployed_model_id)
   
   # The predictions are a google.protobuf.Value representation of the model's predictions.
   predictions = response.predictions
   for prediction in predictions:
      print(" prediction:", dict(prediction))
   return predictions

当我点击我的云函数时，它会下载它并将其转换为该形状，但随后我收到此错误消息和堆栈跟踪（图像具有白色背景，因此 [255, 255, 255] 是正确的）：

File "/env/local/lib/python3.7/site-packages/google/cloud/aiplatform_v1/services/prediction_service/client.py", line 436, in predict
    request.instances.extend(instances)
  File "/env/lib/python3.7/_collections_abc.py", line 990, in extend
    self.append(v)
  File "/env/lib/python3.7/_collections_abc.py", line 971, in append
    self.insert(len(self), value)
  File "/env/local/lib/python3.7/site-packages/proto/marshal/collections/repeated.py", line 177, in insert
    pb_value = self._marshal.to_proto(self._pb_type, value)
  File "/env/local/lib/python3.7/site-packages/proto/marshal/marshal.py", line 208, in to_proto
    pb_value = rule.to_proto(value)
  File "/env/local/lib/python3.7/site-packages/google/cloud/aiplatform/helpers/_decorators.py", line 33, in to_proto
    return super().to_proto(value)
  File "/env/local/lib/python3.7/site-packages/proto/marshal/rules/struct.py", line 80, in to_proto
    raise ValueError("Unable to coerce value: %r" % value)
**ValueError: Unable to coerce value:** <tf.Tensor: shape=(1, 180, 180, 3), dtype=float32, numpy=
array([[[[255., 255., 255.],
         [255., 255., 255.],
         [255., 255., 255.],
         ...,
         **snip**
         ...,
         [255., 255., 255.],
         [255., 255., 255.],
         [255., 255., 255.]]]], dtype=float32)>

我应该如何格式化对模型的实际调用？

4

0 回答 0