ResourceExhaustedError in train_c3d_ucf101.py

[Wei-i]

I am sorry to disturb you, authors. I guess that this error is probably caused by my poor GPU configuration(NVIDIA GTX960M 2G GDDR5).
Which codes can I change or optimize to suit my poor computer? I am sincerely looking forward for your answer, thanks!

[rocksyne]

Try this tool from the tensorflow library.

tf.data.Dataset.from_generator

It will help use a generator function to load data in batches. I am providing a pseudocode example below.

# now the generator function
def data_generator(original_data:list=None,batch_size:int=10)->list:

	# make sure the input data is not none
	if original_data is None:
		sys.exit("Input data invalid!")

	#we need to run this forever
	while 1:

		# run a for loop to be able to fetch the data in batches
		for batch in range(0,len(original_data),batch_size):
			data = original_data[batch:batch+batch_size] # fetch bacthes
			
			# I am not sure if this should be a numpy array
			# but if it should, please use the appropriate type casting
			# data = np.array(data).astype(np.float32) --> this is an example
			# Find your appropriate casting and set it as such

			# yield this current batch of data
			yield data


"""
H0w to use this in your code
-------------------------------------------------

import tensorflow as tf
from data_generator import data_generator

custom_genenerator = data_generator(parameter_1, optional_param_2)

dataset = tf.data.Dataset.from_generator(custom_genenerator,(tf.float32, tf.int16))
iterator = dataset.make_one_shot_iterator()

x,y = iterator.get_next()



Check the following references for explanation
--------------------------------------------------
Ref: 
https://sknadig.me/TensorFlow2.0-dataset/
https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_generator

"""

[Wei-i]

Try this tool from the tensorflow library.

tf.data.Dataset.from_generator

It will help use a generator function to load data in batches. I am providing a pseudocode example below.

# now the generator function
def data_generator(original_data:list=None,batch_size:int=10)->list:

	# make sure the input data is not none
	if original_data is None:
		sys.exit("Input data invalid!")

	#we need to run this forever
	while 1:

		# run a for loop to be able to fetch the data in batches
		for batch in range(0,len(original_data),batch_size):
			data = original_data[batch:batch+batch_size] # fetch bacthes
			
			# I am not sure if this should be a numpy array
			# but if it should, please use the appropriate type casting
			# data = np.array(data).astype(np.float32) --> this is an example
			# Find your appropriate casting and set it as such

			# yield this current batch of data
			yield data


"""
H0w to use this in your code
-------------------------------------------------

import tensorflow as tf
from data_generator import data_generator

custom_genenerator = data_generator(parameter_1, optional_param_2)

dataset = tf.data.Dataset.from_generator(custom_genenerator,(tf.float32, tf.int16))
iterator = dataset.make_one_shot_iterator()

x,y = iterator.get_next()



Check the following references for explanation
--------------------------------------------------
Ref: 
https://sknadig.me/TensorFlow2.0-dataset/
https://www.tensorflow.org/api_docs/python/tf/data/Dataset#from_generator

"""

Thanks for replying！But I am sorry I have not made progress yet.
Are there any codes about batch size that can be modified to reduce directly？