How do you train DRN-D-22?

Question

How do you train DRN-D-22?

acgtyrant opened this issue 6 years ago · comments

I trained it with python3 segment.py train -d <data_folder> -c <category_number> -s 896 --arch drn_d_22 --batch-size 14 --epochs 250 --lr 0.001 --momentum 0.99 --step 100, and multi-scale test only get 66.49.

It seems that you have many GPUs, you may use a huge batch-size or big crop size, right?

Isaac Ge · Answer 1 · Thu May 10 2018 10:41:16 GMT+0800 (China Standard Time)

OK, I forget to use --bn-sync.

Isaac Ge · Answer 2 · Thu May 10 2018 16:44:12 GMT+0800 (China Standard Time)

I add --bn-sync and it overflow in epoch 3, it comfirms that the implementation of synced bn is not correct.

Jakub Langr · Answer 3 · Wed Dec 25 2019 05:24:55 GMT+0800 (China Standard Time)

Hi @acgtyrant would you be able to upload the models to e.g. dropbox as they don't seem to be accessible on the Princeton site? That'd be awesome! : 👏 😎