• rollouts.compute_returns(next_value, args.use_gae, args.gamma, args.gae_lambda)
  • next_value=0
• return torch.cat(rnn_hidden_states_new, dim=-1) in policy.py

python -u iBatchLearn.py --gpuid 1 --repeat 1 --incremental_class --optimizer SGD \
    --n_permutation 1 --force_out_dim 12 --schedule 30 --batch_size 32 --model_name MLP1000 \
    --agent_type regularization --agent_name SI --lr 0.0001 --reg_coef 0.3

python -u iBatchLearn.py --gpuid 1 --repeat 1 --incremental_class --optimizer SGD \
    --n_permutation 1 --force_out_dim 12 --schedule 30 --batch_size 32 --model_name MLP1000 \
    --agent_type regularization --agent_name L2 --lr 0.0001 --reg_coef 0.3

# Imagenet
CUDA_VISIBLE_DEVICES=0 python imagenet.py /ssd2/chenwy/imagenet_final --resume visda_src_model.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=0 python imagenet.py /ssd2/chenwy/imagenet_final --resume visda_src_model.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=1 python imagenet.py /ssd2/chenwy/imagenet_final --resume visda_src_model.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=1 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/metatrain_policy.reinforce.sgd.gamma0_LR1e4.lambda0.1_lrmeta2.nodecay_all_hidden20_loss.avg_epoch50.exp.earlystop1_min10iter_step3_win.shrink.20/model_best.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=0 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=2 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR5e-5_all_1to1_epoch30/epoch_22_checkpoint.pth.tar --evaluate
CUDA_VISIBLE_DEVICES=1 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/metatrain_policy.reinforce.sgd.gamma0_LR1e4.lambda0.1_lrmeta2.nodecay_all_hidden20_loss.avg_epoch50.exp.earlystop1_min10iter_step3_win.shrink.20/epoch_30_checkpoint.pth.tar --evaluate
python imagenet.py /ssd2/chenwy/imagenet_final --resume visda_src_model.pth.tar --evaluate --gpu 1
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/metatrain_policy.reinforce.sgd.gamma0_LR1e4.lambda0.1_lrmeta2.nodecay_all_hidden20_loss.avg_epoch50.exp.earlystop1_min10iter_step3_win.shrink.20/epoch_30_checkpoint.pth.tar --evaluate --gpu 2
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR5e-5_all_1to1_epoch30/epoch_22_checkpoint.pth.tar --evaluate --gpu 4
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_fconly_epoch30/epoch_27_checkpoint.pth.tar --evaluate --gpu 5
python imagenet.py /ssd2/chenwy/imagenet_final --resume runs/Res101_Eval_Mean_Poly_LR3e-4_all_1to0.5_epoch30/epoch_30_checkpoint.pth.tar --evaluate --gpu 6
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR4e-4_all_1to1_lwf0.1_epoch30/epoch_30_checkpoint.pth.tar --evaluate --gpu 1
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR5e-5_all_1to1_lwf0.1_epoch30/epoch_26_checkpoint.pth.tar --evaluate --gpu 2
python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-4_all_1to0.5_lwf0.1_epoch30_2/epoch_23_checkpoint.pth.tar --evaluate --gpu 3
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
CUDA_VISIBLE_DEVICES=3 python imagenet.py /ssd2/chenwy/imagenet_final --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/epoch_25_checkpoint.pth.tar --evaluate --gpu 3
### Visda17 source
python visda17_source.py --resume ./runs/Res101_Eval_Mean_Poly_LR1e-3_all_1to1_epoch30/model_best.pth.tar --evaluate --gpu 1 # 99.633
python visda17_source.py --resume ./runs/Res101_Eval_Mean_Poly_LR4e-4_all_1to1_lwf0.1_epoch30/model_best.pth.tar --evaluate --gpu 0 # 99.986
python visda17_source.py --resume ./runs/Res101_Eval_Mean_Poly_LR3e-4_all_1to0.5_epoch30/model_best.pth.tar --evaluate --gpu 7 # 99.585
python visda17_source.py --resume ./runs/Res101_Eval_Mean_Poly_LR1e-4_all_1to0.5_lwf0.1_epoch30_2/model_best.pth.tar --evaluate --gpu 6 # 99.977
python visda17_source.py --resume ./runs/metatrain_policy.reinforce.sgd.gamma0_LR1e4.lambda0.1_lrmeta2.nodecay_all_hidden20_loss.avg_epoch50.exp.earlystop1_min10iter_step3_win.shrink.20/model_best.pth.tar --evaluate --gpu 0 # 99.974

ASG

Project | Paper

Automated Synthetic-to-Real Generalization.
Wuyang Chen, Zhiding Yu, Zhangyang Wang, Ming-Yu Liu, Yong Jae Lee, Alexander G. Schwing, Jan Kautz, Anima Anandkumar.
In ICML 2020.

More details coming soon...

Citation

If you use this code for your research, please cite:

@incollection{chen2020automated,
 author = {Chen, Wuyang and Yu, Zhiding and Wang, Zhangyang and Anandkumar, Anima},
 booktitle = {Proceedings of Machine Learning and Systems 2020},
 pages = {8272--8282},
 title = {Automated Synthetic-to-Real Generalization},
 year = {2020}
}