search

zyxu1996 / Efficient-Transformer

Online !!! Application of an efficient transformer improved based on Swin transformer on remote sensing segmentation
Apache License 2.0
74 stars 7 forks source link

comparision of resultant images with different batchsizes #1

Closed Syeda-Farhat closed 11 months ago

Syeda-Farhat commented 2 years ago

The resultant image changes with different batchsizes, why?

problem

)

zyxu1996 commented 2 years ago

How did you run the test code command? The result should not be related to the batch size. Maybe you should run the test code in the corresponding work_dir.

zyxu1996 commented 2 years ago

It seems the network params are random initialized.

zyxu1996 commented 2 years ago

The val batchsizes are used to accelerate the inferring process.

Syeda-Farhat commented 2 years ago

I just checked with different batch sizes and in the same corresponding work_dir. yes you are right results should not be affected but unfortunately :(

p1
zyxu1996 commented 2 years ago

I noticed that your trained weights are saved in tmp_save dir, but the information in the test command is num1. Both the names should be same, otherwise the model can not read the right trained weights.

zyxu1996 commented 2 years ago

Have you trained the model? If you don't train the model, you will not get the trained weights in the work_dir, so the loaded trained weight are randomly initialized.

zyxu1996 commented 2 years ago

Different loaded weight will output different results.

Syeda-Farhat commented 2 years ago

oh yes, now I get the point. I will try again with these changes. Thank You so much for your response.