kgl-prml / contrastive-adaptation-network-for-unsupervised-domain-adaptation Goto Github PK

When I use the test method and encounter this problem, I would like to ask you how to solve it，The corresponding weight folder does not correspond after training
FileMotFoundError: No such file or directory: './experiments/ckpt/$lexperiment_name}/ckpt * .weights

Hi!
I'm trying to reproduce results on Office-31.
I've trained each experiment 10 times. I've selected the best results in each train(not the last one). I've calculated average of 10 samples and it is worse in D->A: in paper result is 78, my result is 77,
in D->W: in paper result is 99.1, my is 98.3 and W->A: in paper result is 77, my result is 75.6.
Can you help me with this issue?

Dear Author,
I appreciate your contribution on the paper CAN(Contrastive Adaptation for UDA).
please, i'm trying to run your github code,but no 'category.txt) for both datasets.
Cloud you suggest me,please.

Hello, could you share the information about the general running time of each experiment in the paper?

Thanks.

the original paper's result is 87.2, but my average result (the best value of 10 experiment results) is 86.6.

Where is the pesudeo label of the target domain used in the cdd.py?

Hi,

Do you know what is the effect of domain specific batch norm on the final accuracy? What would be the drop without it?

Hi, thanks for your code!

I am wondering how to split the training and testing sets on both Office-31 and VisDA-2017 datasets in your work?
I could not find the related code.
Could you give me any suggestions?
Thanks a lot!

Thanks for the excellent work and the highly-organized code.
I'm a little confused about the BN behavior of FC for CAN. According to what i observed, during the training, the target domain's FC BN was not updated. Only the source domain's FC BN layer was updated through the CE loss.
When testing, however, the target domain's FC BN was used. Would this harm the performance of the network? Since the target domain's FC BN haven't been trained and may introduce biases.
Thank you!

Dear Author,
I appreciate your contribution on the paper CAN(Contrastive Adaptation for UDA).
please, i'm trying to run your github code,but no 'category.txt) for both datasets.
Cloud you suggest me,please.

Hi, thank you for sharing your great work :)
I am struggling to debug the code since your code uses multi-processing in dataloader.
When I use debug mode in pycharm, the error occurs at the dataloader (it's okay with the simple run in pycharm).
I attached the error log below.
Thank you!

for sample in iter(dataloader):

File "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py", line 571, in next
self._shutdown_workers()
File "/usr/local/lib/python3.7/dist-packages/torch/utils/data/dataloader.py", line 659, in _shutdown_workers
w.join()
File "/usr/lib/python3.7/multiprocessing/process.py", line 140, in join
res = self._popen.wait(timeout)
File "/usr/lib/python3.7/multiprocessing/popen_fork.py", line 48, in wait
return self.poll(os.WNOHANG if timeout == 0.0 else 0)
File "/usr/lib/python3.7/multiprocessing/popen_fork.py", line 28, in poll
pid, sts = os.waitpid(self.pid, flag)
KeyboardInterrupt

Thank you for sharing code!

I have a question about the function get_centers that computes the centers of clusters.
Did you forget to normalize features before multiply them per mask? Or simply the normalization is in the net forward?

Hi, authors!
Thanks for sharing the code and the interesting work! I am confused about the Class-aware Sampling (CAS). May I ask how CAS samples data from the target domain? I would be grateful for any suggestions on this.

Best wishes.

Hi,

I just tested your code on Office-Home and ImageCLEF-DA. The performance is not good. Any suggestions on how to solve this? Perhaps your proposed method is overfitting on VisDA and Office-31.

The original paper only reports resnet-101 result for Vis-DA dataset.
Could you please add an extra result of resnet-50 for Vis-DA dataset?
Because really lots of papers are doing RseNet-50 on Vis-DA!!!

Hi!
can you please join the script you used in order to organize VisDA2017 in the way you mention?
the original VisDA2017 does not have a category.txt file
Thanks!