bo-zhang-cs / cacnet-pytorch Goto Github PK

Thanks for providing the implementation. I have a query, in the steps you first mention 'python train_composition_classification.py' which saves the checkpoint in 'best_acc.pth'. Next, we run 'python train_image_cropping.py' which again saves different checkpoint. I can see these steps should be performed sequentially. So how is the checkpoint saved in best_acc.pth is used in the next module for image cropping? The script does not call for the previous weights to be loaded anywhere. Please help to clarify this. Thanks.

for i in range(len(P_w)):
for j in range(len(P_h)):
anchors[k, 1] = P_w[j]
anchors[k, 0] = P_h[i]
k += 1
return anchors
TypeError: can't assign a numpy.int32 to a Variable[CPUType]

How to deal with it

Hi, thanks for your sharing awesome codes.

Dose your reproduced CACNet only normally run with the setting of keep_aspect_ratio=False? I have tried to run with the setting of keep_aspect_ratio=True and crop_batch_size=1, however there encountered the following error, I want to know what cause it. Could you please share your perspectives about keep_aspect_ratio=True? :-)

File "/root/CACNet/CACNet.py", line 194, in forward
    regression[...,0::2] = offsets[..., 0::2] + self.all_anchors[...,0:1]
RuntimeError: The size of tensor a (44) must match the size of tensor b (28) at non-singleton dimension 2

How to run the test file and save the cropped image

Hello, can you share the original basic code?

Do you have any sample code to use pre-trained model to crop a single image input, or any suggestion ?