test.py

[waquey]

Is there any test code we can try to inference?

After training, I've tried to assign placeholder and inference.

However, it'll get the following errors:

tensorflow.python.framework.errors_impl.FailedPreconditionError: Attempting to use uninitialized value layer2/conv2d/kernel [[node layer2/conv2d/kernel/read (defined at test.py:967) ]]

===================== Code ========================= input_name ='image_tensor'

t_input = tf.compat.v1.placeholder(
    shape=[1, 128, 128, 3], dtype=tf.float32,
    name=input_name)

saver = tf.compat.v1.train.import_meta_graph(input_checkpoint + '.meta', 
                           clear_devices=True,
                           input_map={'Cast:0': t_input})
graph = tf.compat.v1.get_default_graph()
input_operation = graph.get_operation_by_name(input_name)

## define output node 
output_name = 'layer5/final_dense'
output_operation = graph.get_operation_by_name(output_name)

.... assign 1 image as cropped_ims, which is 1x128x128x3 ... 

with tf.compat.v1.Session(graph=graph) as sess: ONet_point_results = sess.run([ output_operation.outputs[0] ], {input_operation.outputs[0]: cropped_ims})

[yinguobing]

I have not figure out how to write a proper test code. For now you need to export the model for inference.

[zhouqianyu0918]

您好，我想请问一下，我使用新的数据集训练网络，那么训练完成后怎么对它进行测试呢？请问有测试的代码吗？

[yinguobing]

@zhouqianyu0918 先实现推演，这是测试的基础。至于测试，它需要有明确的测试数据，测试指标等等。可以尝试在论文里找找通用的测试协议。