I'm using inception_client.py to perform predictions with tensorflow serving and it works very well. But when I'm trying to batch the images on request and execute, tensorflow serving API raises this exception:
AbortionError(code=StatusCode.INVALID_ARGUMENT, details="Input to reshape is a tensor with 20480 values, but the requested shape has 2048 [[Node: pool_3/_reshape = Reshape[T=DT_FLOAT, Tshape=DT_INT32, _output_shapes=[[1,2048]], _device="/job:localhost/replica:0/task:0/device:CPU:0"](pool_3, pool_3/_reshape/shape)]]")
As a pre-process, I've readed two images, converted to tensor, normalize and create a list to use on make_tensor_proto:
proto = tf.contrib.util.make_tensor_proto(images, shape=tf.shape(images).numpy().tolist())
request.inputs['image'].CopyFrom(proto)
result = stub.Predict(request, 120.0)
I've created this model using retrain.py, retraining InceptionV3 model
Anybody can help to understand this problem?