Dataset files missing in submodule applications/dnn/data

[jose-cubero]

Hello,

I tried to run run_inference.py, after setting task = "imagenet" in the config file.
The script failed during initialization due to a non-existing data-set file.
Error message: No such file or directory: '../data/datasets/imagenet/x_val_MLperfRN50_1000.npy'.

I checked the submodule cross-sim-data and could not locate the file.
I noticed that many other files referenced in applications/dnn/dataset_loaders.py are also missing.

Can you please make the required npy files available in the submodule(s) or provide instructions on generating them?

Thanks and regards,
-José

[bfeinberg]

José,
Due to the licensing restrictions on the ImageNet dataset we cannot distribute it directly. You'll need to request access from (https://www.image-net.org/) to be able to download it.

Once you have a copy you can use the scripts in the imagenet_preprocess directory (https://github.com/sandialabs/cross-sim/tree/main/applications/dnn/inference/helpers/imagenet_preprocess) to create the binary file those scripts are expecting.
Let us know if you have any issues,
--Ben