search

jhhuang96 / ConvLSTM-PyTorch

ConvLSTM/ConvGRU (Encoder-Decoder) with PyTorch on Moving-MNIST
MIT License
383 stars 85 forks source link

After training for 20 epochs with default params, the validation results seem to be quite blurry. #10

Open Crestina2001 opened 4 months ago

Crestina2001 commented 4 months ago

image Could anyone offer a solution. Many thanks.

Crestina2001 commented 4 months ago

The config is like:

convlstm: False convgru: False batch_size: 4 lr: 0.0001 frames_input: 10 frames_output: 10 epochs: 20

When convlstm and convgru are both False, it actually uses convgru.

Crestina2001 commented 4 months ago

Here is the result using convlstm(epochs: 20): image

Crestina2001 commented 4 months ago

After training for 150 epochs, the result seem to improve, but not by much. I get the best val loss when epoch = 144. Here is the visualization: validation_sample_3 It is still far from satisfaction.

zxx-push commented 1 week ago

guru,hello!Could you share your code with me.Great thanks!