Giters

chenxin061 / pdarts

Codes for our paper "Progressive Differentiable Architecture Search:Bridging the Depth Gap between Search and Evaluation"

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Problem while running test.py

Catosine opened this issue · comments

commented

Hello,

Thanks for your amazing work! It is brilliant!
However, I have a problem when trying to search a model for a face dataset. The search is successful but I am currently stuck at testing.
When I try with test.py with some minor modifications (i.e. load dataset from my own ones), it always turns out with a shape error: the weight.pt does not have exactly the same shape as the model suggests.
I was wondering if I need to change other parts of the testing codes, or I may be wrong during training. (FYI, I use the same genotypes to train on different dataset.)

Thanks!

commented

To be more specific, what does "--init_channels" do in test.py? Should that be the same as my data?
(My data shape: (3, 112, 112))

Also, I was wondering if the size issue is caused by using different layers in model searching and testing.

commented

Update: It seems that NetworkCIFAR can only work with CIFRA10/100. I tried the test.py with pretrained CIFRAR10.pt on CIRFAR10 and my own dataset. It works fine on the first set but has the same size mismatch issue on the latter one.

commented

Update: It seems that NetworkCIFAR can only work with CIFRA10/100. I tried the test.py with pretrained CIFRAR10.pt on CIRFAR10 and my own dataset. It works fine on the first set but has the same size mismatch issue on the latter one.

I have encountered the same problem. How can I solve it?
python test.py \ --auxiliary \ --model_path log_pathsearch-note_of_this_run-20190615-081053/weights.pt \ --data /path/to/your/data

commented

Update: It seems that NetworkCIFAR can only work with CIFRA10/100. I tried the test.py with pretrained CIFRAR10.pt on CIRFAR10 and my own dataset. It works fine on the first set but has the same size mismatch issue on the latter one.

I have encountered the same problem. How can I solve it?
python test.py \ --auxiliary \ --model_path log_pathsearch-note_of_this_run-20190615-081053/weights.pt \ --data /path/to/your/data

Here is the thing: as far as I understand, test.py is only used to see the performance of a pretrained model. It is NOT used to retrain your searched structures. You may look for train_cifar.py and train_imagenet.py to retrain any pretrained models.

Best,