agaldran / cost_sensitive_loss_classification Goto Github PK

Hi
thank you for this git, it is handy and well-written!

I have a question about the final cost matrix you define - why would we want values different than 0 on the diagonal?
let's say we have 3 classes, why would the M_00, M_11, M_22 be different than 0? this is the correct classification loss...

I understand the concept of the averaging: M = 0.5 * M + 0.5 * (1-M_opht)
But after defining this matrix M, why shouldn't we set all the diagonals to zero?
thanks!

python train.py --csv_train train.csv --model_name resnext50 --base_loss gls --labmd 10 --exp 2 --save_path gls_reg_1e2_resnext50_exp2

Should be:
python train.py --csv_train train.csv --model_name resnext50 --base_loss gls --lambd 10 --exp 2 --save_path gls_reg_1e2_resnext50_exp2

when I run the script prepare_test_data.py

this line 18 occurs error
from utils.image_preprocessing import correct_illumination

can you please provide the script image_preprocessing.py which include the function correct_illumination?

thanks very much.