search

kmsiapps / e2e-dnn-comm-for-image

End to end deep neural network based semantic communication system for image.
58 stars 6 forks source link

The experimental environment #4

Open Caowintay opened 1 year ago

Caowintay commented 1 year ago

Dear author, I recently set up a lab environment and had some problems with the version, can you provide a specific version of the experiment? Such as the version of CUDNN, CUDA and Tensorflow. Thanks a lot.

kmsiapps commented 1 year ago

Can you explain more about the issue? I used the official tensorflow docker image (https://www.tensorflow.org/install/docker) in the Ubuntu 20.04 environment.

Caowintay commented 1 year ago

Can you explain more about the issue? I used the official tensorflow docker image (https://www.tensorflow.org/install/docker) in the Ubuntu 20.04 environment.

Thanks a lot for your replying. I have run the code. But now I have a new problem. Could you give me some help? The problem is: I ran the code with Dataset CIFAR10, after training, I ran evaluate.py, but the results are not ideal. Such as followings. The reconstructed image is black. The train and test loss is about 0.3. image

kmsiapps commented 1 year ago

That's weird. Are the image pixel values between 0 and 1? Inconsistent image pixel values may cause those problems. Also, the loss value of 0.3 is too high. You may check the training procedure once again.

Caowintay commented 1 year ago

That's weird. Are the image pixel values between 0 and 1? Inconsistent image pixel values may cause those problems. Also, the loss value of 0.3 is too high. You may check the training procedure once again.

Ok,Thanks a lot. I will try again.