consult training hyperparameters

[shaofeifei11]

Dear Sir or Madam,

First of all thank you very much for your excellent work.

I want to follow your work, but I want to consult some training hyperparameters.

Could you give me some training hyperparameters for other models and datasets based on the example training parameters below? (CUB+vgg16_CAM, CUB+InceptionV3_CAM, CUB+resnet50_CAM, Imagenet+vgg16_CAM, Imagenet+InceptionV3_CAM, Imagenet+resnet50_CAM)

python main.py --dataset_name CUB \
               --architecture vgg16 \
               --wsol_method cam \
               --experiment_name CUB_vgg16_CAM \
               --pretrained TRUE \
               --num_val_sample_per_class 5 \
               --large_feature_map FALSE \
               --batch_size 32 \
               --epochs 50 \
               --lr 0.00001268269 \
               --lr_decay_frequency 15 \
               --weight_decay 5.00E-04 \
               --override_cache FALSE \
               --workers 4 \
               --box_v2_metric True \
               --iou_threshold_list 30 50 70 \
               --eval_checkpoint_type last
               --norm_method ivr

Best regards

[SanghyukChun]

@shaofeifei11 Hi, can this issue (#22) help you?

[shaofeifei11]

Thank you for your timely reply! Yes, this is what i want. Thank you!

[shaofeifei11]

Could you tell the resize value of InceptionV3 and resnet50?

[SanghyukChun]

What do you mean for resize? If you mean the input image size, all inputs are the same as 224 https://github.com/clovaai/wsolevaluation/blob/96f4b88679af41f005824661f47e6fe3628af2aa/inference.py#L35

We only searched the following hyperparameters for each method.

If you mean the feature map size, large_feature_map TRUE denotes 28x28, FALSE detnoes 14x14 https://github.com/clovaai/wsolevaluation/blob/96f4b88679af41f005824661f47e6fe3628af2aa/wsol/vgg.py#L288