Why there is no GaussianBlur in test transform?

Data Purchasing Challenge 2022
#1

it reduce my f1-score by 0.16(val_dataset)
i think this is necessary

I also hope to remove the following code in model.py

for param in self.base_model.features.parameters():
param.requires_grad = False

f1 add GaussianBlur remove Freeze feature extration layers
0.226 x x
0.394 √ x
0.569 √ √
#2

does gaussianblur even make sense with the small particles ? Someone should look at how an image with this applied looks like. Let’s assume it totally washes out the small particles, but still is recognizable, but it just looks different, this could explain the worse scores in eval. Or maybe it does only affect the speed of convergence.

#3

Hi @chuifeng

Yes the GaussianBlur in training should have been applied with a probability value during training to get higher scores. Or as you suggest, could be used during test which will also increase the score.

Though I’d still like to stress that the gap in scores between random purchase and the all-label purchase is the criteria that is more important that the raw scores.

You can refer to the discussion here:

#4

thanks for your reply!
I raise this question because I am worried that this may lead to this situation: the better the samples selected, the lower the F1-score
for GaussianBlur will change the feature of images.
If officials have confirmed that the above problems will not occur， that is great

1 Like