forked from akanazawa/hmr
-
Notifications
You must be signed in to change notification settings - Fork 0
/
do_train.sh
executable file
·13 lines (10 loc) · 879 Bytes
/
do_train.sh
1
2
3
4
5
6
7
8
9
10
11
12
13
# TODO: Replace with where you downloaded your resnet_v2_50.
PRETRAINED=/scratch1/projects/tf_datasets/models/resnet_v2_50/resnet_v2_50.ckpt
# TODO: Replace with where you generated tf_record!
DATA_DIR=/scratch1/storage/hmr_release_files/test_tf_datasets/
CMD="python -m src.main --d_lr 1e-4 --e_lr 1e-5 --log_img_step 1000 --pretrained_model_path=${PRETRAINED} --data_dir ${DATA_DIR} --e_loss_weight 60. --batch_size=64 --use_3d_label True --e_3d_weight 60. --datasets lsp,lsp_ext,mpii,h36m,coco,mpi_inf_3dhp --epoch 75 --log_dir logs"
# To pick up training/training from a previous model, set LP
# LP='logs/<WITH_YOUR_TRAINED_MODEL>'
# CMD="python -m src.main --d_lr 1e-4 --e_lr 1e-5 --log_img_step 1000 --load_path=${LP} --e_loss_weight 60. --batch_size=64 --use_3d_label True --e_3d_weight 60. --datasets lsp lsp_ext mpii h36m coco mpi_inf_3dhp --epoch 75"
echo $CMD
$CMD