| #!/bin/bash |
| |
| |
| method="$1" |
| task=$(python3 config.py --print_task) |
| case "${task}" in |
| 'DIS5K') epochs=200 && val_last=50 && step=5 ;; |
| 'COD') epochs=150 && val_last=50 && step=5 ;; |
| 'HRSOD') epochs=150 && val_last=50 && step=5 ;; |
| 'General') epochs=250 && val_last=50 && step=2 ;; |
| 'General-2K') epochs=250 && val_last=30 && step=2 ;; |
| 'Matting') epochs=100 && val_last=30 && step=2 ;; |
| esac |
| testsets=NO |
| |
|
|
| |
| devices=$2 |
| nproc_per_node=$(echo ${devices%%,} | grep -o "," | wc -l) |
|
|
| to_be_distributed=`echo ${nproc_per_node} | awk '{if($e > 0) print "True"; else print "False";}'` |
|
|
| echo Training started at $(date) |
| if [ ${to_be_distributed} == "True" ] |
| then |
| |
| echo "Multi-GPU mode received..." |
| CUDA_VISIBLE_DEVICES=${devices} \ |
| torchrun --standalone --nproc_per_node $((nproc_per_node+1)) \ |
| train.py --ckpt_dir ckpt/${method} --epochs ${epochs} \ |
| --testsets ${testsets} \ |
| --dist ${to_be_distributed} \ |
| --resume xx/xx-epoch_244.pth \ |
| |
| else |
| echo "Single-GPU mode received..." |
| CUDA_VISIBLE_DEVICES=${devices} \ |
| python train.py --ckpt_dir ckpt/${method} --epochs ${epochs} \ |
| --testsets ${testsets} \ |
| --dist ${to_be_distributed} \ |
| --resume xx/xx-epoch_244.pth |
| |
| fi |
|
|
| echo Training finished at $(date) |
|
|