search

lizhengwei1992 / Semantic_Human_Matting

Semantic Human Matting
527 stars 145 forks source link

random_scale_and_creat_patch #21

Open wwq66 opened 4 years ago

wwq66 commented 4 years ago

Why should randomly scale and create path of input images? I do not understand why create pathes with randomly selecting areas in the image.

Jason-xin commented 4 years ago

When I run ./test_camera.sh (--without_gpu), this error occurred: AttributeError: Can't get attribute 'RD_FPNnet' on <module 'model.T_Net' from '/Users/jasonxin/Documents/dlworkspace/Semantic_Human_Matting/model/T_Net.py'> My torch version is 1.0.0.

wwq66 commented 4 years ago

When I run ./test_camera.sh (--without_gpu), this error occurred: AttributeError: Can't get attribute 'RD_FPNnet' on <module 'model.T_Net' from '/Users/jasonxin/Documents/dlworkspace/Semantic_Human_Matting/model/T_Net.py'> My torch version is 1.0.0.

Suggest that run the code with pytorch 0.4.0 or you will get mistakes.

Jason-xin commented 4 years ago

When I run ./test_camera.sh (--without_gpu), this error occurred: AttributeError: Can't get attribute 'RD_FPNnet' on <module 'model.T_Net' from '/Users/jasonxin/Documents/dlworkspace/Semantic_Human_Matting/model/T_Net.py'> My torch version is 1.0.0.

Suggest that run the code with pytorch 0.4.0 or you will get mistakes.

@cqww I have changed it to the pytorch 0.4.0, but the same errors...

xiankgx commented 4 years ago

Why should randomly scale and create path of input images? I do not understand why create pathes with randomly selecting areas in the image.

This is a data augmentation technique for improving the generalization-ability of the model.