Please can someone describe how to properly obtain the ImageNet dataset (to be precise the ImageNet 2012 Classification Dataset).
What I attempted so far
The ImageNet webpage refers the user to download the ImageNet dataset from Kaggle. However, the Kaggle webpage it refers belongs to the Image Localization (not classification) challenge.
I have also requested a download from the ImageNet webpage which is pending since almost one year.
These are the detailed steps on how I obtained ImageNet and ran a PyTorch example training on it:
1. Go to https://www.image-net.org/download.php
2. Request to download ImageNet
3. Wait about 5 days for approval, write to them if the waiting period is over.
4. [I think you can skip this step] Download the Development Kit from the ILSVRC2017 page
5. Download the images from the ILSVRC2012 page
a. Training images (Task 1 & 2) 138 GB
b. Validation images (all tasks) 6.3 GB
c. Test images (all tasks) 13 GB
6. [I think you can skip this step if you use the script from step 8!] Unpack the tar files
a. mkdir val
b. tar -C val/ -xvf ILSVRC2012_img_val*.tar
c. mkdir test
d. tar -C test/ -xvf ILSVRC2012_img_test_v10102019.tar
e. media train
f. tar -C train/ -xvf ILSVRC2012_img_train.tar
7. Confirm the number of images in each folder
a. ls val/ | wc -l # should give 50,000
b. ls test/ | wc -l # should give 50,000
8. Run the script extract_ILSVRC.sh from the PyTorch GitHub [https://github.com/pytorch/examples/blob/main/imagenet/extract_ILSVRC.sh]
# imagenet/train/
# ├── n01440764
# │ ├── n01440764_10026.JPEG
# │ ├── n01440764_10027.JPEG
# │ ├── ......
# ├── ......
# imagenet/val/
# ├── n01440764
# │ ├── ILSVRC2012_val_00000293.JPEG
# │ ├── ILSVRC2012_val_00002138.JPEG
# │ ├── ......
# ├── ......
9. Run a PyTorch example training on your ImageNet dataset [e.g. from the PyTorch examples GitHub repository https://github.com/pytorch/examples/blob/main/imagenet/main.py]
ImageNet is available in torchvision datasets. https://pytorch.org/vision/stable/generated/torchvision.datasets.ImageNet.html
