tim-learn / shot-plus Goto Github PK

Would you please provide the commands used to reproduce the results on VisDA-C dataset? Your paper stated that the result for SHOT-IM++ and SHOT++ is 85.0 and 87.3 respectively, but I could only get 82.8~ by running python image_source.py --output ckps/source/ --da uda --gpu_id 1 --dset VISDA-C --net resnet101 --lr 1e-3 --max_epoch 10 --s 0 and python image_target.py --cls_par 0.3 --da uda --dset VISDA-C --gpu_id 0 --s 0 --output_src ckps/source/ --output ckps/target/ --net resnet101 --lr 1e-3 for training. Thanks.

First of all, thank you for your outstanding work.
I have a question about the performance of Source-only ++(SHOT++), which is superior than Souce-only (SHOT) with a large margin.
But I didn't see the difference with SHOT in the image_source script.
Is this performance evaluted after executing "python image_mixmatch.py --ps 0.0 --cls_par 0.0 --model source --gpu_id $1 --s $s --output_tar "ckps/s"$2 --output "ckps/mm"$2 --seed $2 --dset office --max_epoch 100"? Is this trained using target data?

Where can I see the latest code? I find that there is just readme and license in this zip.