build_imagenet_data.py requires validation images to be in synset directories

Update with requested information

Describe the problem

I want to prepare the ImageNet dataset for training InceptionV3 as described in research/slim/README.md. As I have already downloaded the ImageNet dataset for some previous project I wanted to reuse it by not downloading it again but rather just build the required data format for tensorflow. (On a side note, the download with the provided script does not work, as the URLs seem to be out of date.)

Now the problem is that the script research/slim/datasets/build_imagenet_data.py requires not only the training images but also the validation images to be in the synset directories like that:

data_dir/n01440764/ILSVRC2012_val_00000293.JPEG
data_dir/n01440764/ILSVRC2012_val_00000543.JPEG

But they don't come sorted into these directories when you log into image-net.org and download the validation dataset. The training images are sorted into sub-directories, and for those the script worked (I commented the processing of the validation images). The ground truth labels for the validation images come via the devkit download in a .txt file.

Did anyone else run into this problem? Any thoughts on how to address this apart from writing a script that sorts the validation images into folders based on the ground truth synset labels?

Thanks in advance!

System information

• What is the top-level directory of the model you are using: models/research/slim
• Have I written custom code (as opposed to using a stock example script provided in TensorFlow): No
• OS Platform and Distribution (e.g., Linux Ubuntu 16.04): Linux Ubuntu 18.04
• TensorFlow installed from (source or binary): source
• TensorFlow version (use command below): 1.14.0
• Bazel version (if compiling from source): 0.29.1
• Python version: 3.6.5
• CUDA/cuDNN version: N/A
• GPU model and memory: N/A
• Exact command to reproduce:

cd models/research/slim/datasets
python build_imagenet_data.py --train_directory=PATH_TO_IMAGE_NET/ILSVRC2012_img_train --validation_directory=PATH_TO_IMAGE_NET/ILSVRC2012_img_val --output_directory=PATH_TO_IMAGE_NET/image-net-out --imagenet_metadata_file=imagenet_metadata.txt --labels_file=imagenet_lsvrc_2015_synsets.txt --bounding_box_file=imagenet_2012_bounding_boxes.csv

The imagenet_2012_bounding_boxes.csv file can be created like this:

cd models/research/slim/datasets
python process_bounding_boxes.py PATH_TO_IMAGE_NET/ILSVRC2012_bbox_train_v2 imagenet_lsvrc_2015_synsets.txt | sort > imagenet_2012_bounding_boxes.csv