the recommended version of tensorflow-gpu、keras and numpy to test the code example of "Next-Frame Video Prediction with Convolutional LSTMs"
MysteryNull opened this issue · 5 comments
Issue Type
Others
Source
source
Keras Version
recommended
Custom Code
Yes
OS Platform and Distribution
No response
Python version
3.8
GPU model and memory
No response
Current Behavior?
pip install -r requirements.txt.
But the tensorflow in the requirements.txt can not be installed.
Standalone code to reproduce the issue or tutorial link
1Relevant log output
No response
Hi,
I was able to run the example mentioned above without any error using python 3.10.
Here is the Gist attached for reference https://gist.github.com/sachinprasadhs/479694ef9d6f26cedcc8cdb4c8920aab
It works, but the performance is really bad, how can I improve the performance of the network? Thank you!
As I can see in the above attached Gist is having very less loss.
However, the main intention of the example is to demonstrate the workflow, not to meet any benchmark performance.
In order to get better accuracy, try to increase the number of epochs and you can do some modification to the model structure if needed.
Thank for your advice.
If you don't have other question, could you please close this issue. Thanks!