rawmarshmellows/deep-photo-aesthetics

deep-photo-aesthetics's Issues

about update parameter ?

• first thans for sharing you code !
• there is one question , in your train code :

ignored_params = list(map(id, model.attribute_weights.parameters())) base_params = filter(lambda p: id(p) not in ignored_params, model.parameters())
optimizer = torch.optim.Adam([ {'params': base_params}, {'params': model.attribute_weights.parameters(), 'lr': fc_lr, 'weight_decay': 1e-5} ], lr=fine_tune_lr)

• why you write this way? there is difference like follow :
  optimizer = torch.optim.Adam([ {'params': model.parameters(), 'lr': fc_lr, 'weight_decay': 1e-5} ], lr=fine_tune_lr)

• can you help me ?thanks again !

CAM Maps

The CAM maps generated by your code doesn't match with the CAM maps produced by the paper.

Error while loading the checkpoint

Hi,
I am getting this error while loading the checkpoint:
---> 10 net.load_state_dict(torch.load(f"{save_path}/{checkpoint}", map_location=lambda storage, loc: storage))
...

RuntimeError: unexpected EOF. The file might be corrupted.

RuntimeError: Given groups=1, weight[512, 2048, 1, 1], so expected input[16, 64, 75, 75] to have 2048 channels, but got 64 channels instead

When I run the training program, I got the following runtime error:
RuntimeError: Given groups=1, weight[512, 2048, 1, 1], so expected input[16, 64, 75, 75] to have 2048 channels, but got 64 channels instead
@kevinlu1211 , can you help me, thanks a lot.

How can I use this for my own images?

Hello!

I'm completely new to python, I'm more like a photographer.
I can use the Notebook until the point when I gent first scores of random photos which are provided by you, but how can I get the most precise scores for my own photo?

Like:

A have a picture, and I just want to get the scores. Is there a way to pass something to command line with arguments to get the scores?
I don't want the Python HTML visualizations, just the scores?

Thank you for your reply!

Bruno

pretrained .pth model file load error.

"RuntimeError: unexpected EOF. The file might be corrupted. " for line:
net.load_state_dict(torch.load(os.path.join(save_path, checkpoint)))
Is the .pth file completed? Thanks a lot.

How did you created the dataset? some of the features are having negative values. what do we mean by that?

Labels in train val test .csv files

Hello,
Thanks for sharing your code. I was wondering if the order the labels in csv and dictionary are different or i am missing something ?
In CSV Files the order is like
BalancingElements ColorHarmony Content DoF Light MotionBlur Object Repetition RuleOfThirds Symmetry VividColor score

But in thecode...
BalacingElements ColorHarmony Content DoF Light MotionBlur Object RuleOfThirds VividColor Repetition Symmetry score
The last few are in different order,
so e.g i want to predict Vivid Colors, In dictionary its at 4th last position, but while training from csv i will be giving it a label from 2nd last column.

I am sorry if i can not explain or missunderstood the code ...
Thanks ...

pretrained model

Hi, great work there. I did try the pertained model but it gave an error as
RuntimeError: unexpected EOF. The file might be corrupted.

So I did train the model from scratch and did try new file in the notebook, it gave an error as
UnpicklingError: A load persistent id instruction was encountered,
but no persistent_load function was specified.

What should I do, I am bit confused. Thanks already.