Need help: numpy array to mxnet ndarray is too slow.

[KeyKy]

numpy array to mxnet ndarray is too slow.

start = time.time()
data_ = [mx.nd.array(img_data)]
label_ = [mx.nd.array(label_data)]
end = time.time()
print 't3: {}'.format(end-start)

t3: 0.293313980103. sometimes i get 1.5s. Need help, thanks! It slows down my training!

[madjam]

How big is img_data and label_data?
FYI: this involves allocating memory and copying contents of numpy array into the allocated memory.
How is this step slowing down your training? how often are you doing this copy?

[KeyKy]

@madjam image_data is (128, 3, 224, 224) label_data is (128,)
it needs to load batch of data before doing net training. Forward and backward is much faster than loading data. So i need to reduce the loading data cost.

[Joke09]

For inference, I have the same problem. The client send jpg to server, then the server use cv2 to do resize. When put the image data into the mx.nd.array, it's very slow. And the Utilization of GPU is low too. How to solve it? Thank you!

[summer-via]

I have the same problem, why is the nd.array() function so slow?

[leezu]

You can use the mx.nd.from_numpy(ndarray, zero_copy=True) if you don't mind that the numpy array is invalid afterwards.