JiaxinZhuang / Baseline

Template for supervision task, such as CUB

BASELINE

---

This repo contains some baseline for public datasets.

Requirement

• Python 3.6+

• Cuda 9.2+ or 10.1+

• Package installed requirement.txt

pip install requirement.txt

Directory

Your directory should look like below after git clone.

.
├── data
├── requirement.txt
├── saved
├── scripts
├── src
└── tags

Create directory models and directory logdirs under saved directory.

Dataset.

#Dataset	#Supported	#Train	#Val	#Mean , STD
MNIST	Y	50,000	10,000	[0.131], [0.308]
CIFAR10	Y	50,000	10,000	[0.502, 0.494, 0.461], [0.249, 0.246, 0.263]
CIFAR100	Y	50,000	10,000	[0.505, 0.488, 0.442], [0.267, 0.256, 0.276]
SVHN	Y	73,257	26,032	[0.437, 0.441, 0.470] [0.200, 0.203, 0.199]

Run

Run your scripts in root directory.

bash ./scripts/CUB_000.sh

Logs

You can observe logs in three ways

• Standard output.
• tail -f Logs file, eg. tail -f 000.log under saved/logdirs/000/
• Tensorboard, which also placed under saved/logdirs/000/