Closed RubensZimbres closed 5 years ago
Hi, all training data are tfrecords. You need to get a directory from https://github.com/tensorflow/tpu/blob/master/tools/datasets/imagenet_to_gcs.py or make your training data like that. E.g. gs://cloud-tpu-test-datasets/fake_imagenet
I'm trying to replicate EfficientNet training with ImageNet, but the directory structure is not clear for me. When I extract ImageNet from ImageNet Object Localization Challenge, I get:
ILSVRC/Annotations/CLS-LOC/train/ILSVRC2012_train_0012345.JPEG
containing.xml
filesILSVRC/Data/CLS-LOC/train/ILSVRC2012_train_0012345.JPEG
containing JPEG filesILSVRC/Data/CLS-LOC/val/ILSVRC2012_val_0012345.JPEG
containing JPEG filesWhen I try to train EfficientNet on a Google Cloud TPU, I get an error that the folder
/train-
was not found:What is the directory structure to feed EfficientNet ? Is it somewhat related to this, containing JPEGs ?
If so, where are the dependent variables (target) placed ? Are they the
synset_mapping.txt
?Thanks in advance.
( for more info about my efforts, access: EfficientNet )