History

Xuanyi Dong d9026be4b2 update decompress codes and figures		2019-04-02 17:06:25 +08:00
..
data	init	2019-02-01 01:27:38 +11:00
logs	update logs	2019-04-01 22:24:56 +08:00
classes.txt	init	2019-02-01 01:27:38 +11:00
compress.py	update ZIP for imagenet	2019-04-02 14:58:25 +08:00
decompress.py	update decompress codes and figures	2019-04-02 17:06:25 +08:00
GDAS.png	update decompress codes and figures	2019-04-02 17:06:25 +08:00
Get-PTB-WT2.sh	init	2019-02-01 01:27:38 +11:00
ImageNet-100.txt	init	2019-02-01 02:40:20 +11:00
imagenet-results.png	update decompress codes and figures	2019-04-02 17:06:25 +08:00
README.md	init	2019-02-01 02:40:20 +11:00
split-imagenet.py	init	2019-02-01 02:40:20 +11:00
tiny-imagenet.py	init	2019-02-01 01:27:38 +11:00

README.md

ImageNet

The class names of ImageNet-1K are in classes.txt.

A 100-class subset of ImageNet-1K : ImageNet-100

The class names of ImageNet-100 are in ImageNet-100.txt.

Run python split-imagenet.py will automatically create ImageNet-100 based on the data of ImageNet-1K. By default, we assume the data of ImageNet-1K locates at ~/.torch/ILSVRC2012. If your data is in a different location, you need to modify line-19 and line-20 in split-imagenet.py.

Tiny-ImageNet

The official website is here. Please run python tiny-imagenet.py to generate the correct format of Tiny ImageNet for training.

PTB and WT2

Run bash Get-PTB-WT2.sh to download the data.