kelkelcheng / gc-net-tensorflow Goto Github PK

Hi,

I am trying to build your model again by Tensorflow 2 because I have some issues running the model on my GPU. The only way I found to run the model is by using Docker; however, it is not easy to make it run in realtime with a camera. My question is, can you give me some ideas about the training process you did for your model like, did you used all the FlyingThings dataset and if there some hyperparameter need to tune.

Thanks in advance,

Hi, I would like to know the total number of parameters of your network and does it match with the number given by the original author (3.5M)?

Thank you :)

Hi,

There is an error when running train.py:

OutOfRangeError (see above for traceback): RandomShuffleQueue '_1_shuffle_batch/random_shuffle_queue' is closed and has insufficient elements (requested 1, current size 0)
	 [[node shuffle_batch (defined at ./GC-Net-Tensorflow/util.py:48) ]]

How can I solve the error?

Thanks

I have one question, How many epochs need to train a model?

hi
Do I have to generate .tfrecords files to use pre-trained model， Can I download "saved model " to test the new images? I downloaded the "saved model" folder directly but had an error using test.py:
OutOfRangeError (see above for traceback): RandomShuffleQueue '_5_shuffle_batch_1/random_shuffle_queue' is closed and has insufficient elements (requested 1, current size 0)
[[Node: shuffle_batch_1 = QueueDequeueManyV2[component_types=[DT_FLOAT, DT_FLOAT, DT_FLOAT], timeout_ms=-1, _device="/job:localhost/replica:0/task:0/cpu:0"](shuffle_batch_1/random_shuffle_queue, shuffle_batch_1/n)]]
My computer configuration cannot support training, can i directly test the images？
thanks for your reply, apologize for my ignorance as a beginner

Hi @kelkelcheng,
Please, I have one last question.
Why you have a 'disparity_filter', I couldn't understand the propose of it. In the paper, they just use Softmax at the end.

    logits = tf.nn.softmax(neg)
    disparity_filter = tf.reshape(tf.range(0, d, 1, dtype=tf.float32), [1, 1, d, 1])
    distrib = conv2d(logits, disparity_filter, 1)