When downloading and extracting ImageNet validation images, they are not organized into labeled directories as with training data (i.e., directories starting with ‘n’ for each class label are absent). This differs from the structured format of the training images. Process_dataset.py in pd.sh applies preprocessing operations only to training images, leaving validation images unprocessed.
Questions and Clarification Needed:
Should validation images be placed in labeled subdirectories manually, similar to the training structure?
Is there an expected modification to process_dataset.py to preprocess validation images as well?
Steps to Reproduce:
Download validation data from ImageNet [https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar].
Extract data, and observe that no label directories are created in val.
When downloading and extracting ImageNet validation images, they are not organized into labeled directories as with training data (i.e., directories starting with ‘n’ for each class label are absent). This differs from the structured format of the training images. Process_dataset.py in pd.sh applies preprocessing operations only to training images, leaving validation images unprocessed.
Questions and Clarification Needed:
Should validation images be placed in labeled subdirectories manually, similar to the training structure?
Is there an expected modification to process_dataset.py to preprocess validation images as well?
Steps to Reproduce:
Download validation data from ImageNet [https://image-net.org/data/ILSVRC/2012/ILSVRC2012_img_val.tar].
Extract data, and observe that no label directories are created in val.