Training hyperparmeter for CheXpert dataset

[hankyul2]

Hi, Thank you for sharing your great work.

I am wondering if you can provide some missing hyperparameter for training CheXpert dataset. I summarize training hyperparameters explained in 4.2.1. of DAM paper and denote missing values with ? in below table. If you can provide these missing values, I would appreciate it very much.

CrossEntropy optimization	setting	value
augmentation	RandomAffine(rotation=?, translation=?, scaling=?)
criterion	CE
optimizer	Adam or AdamW (?)
epoch	2
batch	32
lr	1e-5
w.d.	1e-5

AUC Maximization	setting	value
augmentation	RandomAffine(rotation=?, translation=?, scaling=?)
criterion	AUCM_MultiLabel
optimizer	PESG
epoch	2
batch	32
lr	0.1
scheduler	decrease 3 times at 2000, 8000 iter (?)
w.d.	0
gamma	choose between (300, 500, 800) (?)

Thank you again!

[hankyul2]

Hi! I am sorry to ask you without reading your code thoroughly. The augmentation is already applied in dataset shown in below snippets. (link)

def image_augmentation(self, image):
    img_aug = tfs.Compose([tfs.RandomAffine(degrees=(-15, 15), translate=(0.05, 0.05), scale=(0.95, 1.05), fill=128)]) # pytorch 3.7: fillcolor --> fill
    image = img_aug(image)
    return image

Thank you