dr-pato / audio_visual_speech_enhancement

Face Landmark-based Speaker-Independent Audio-Visual Speech Enhancement in Multi-Talker Environments
https://dr-pato.github.io/audio_visual_speech_enhancement/
Apache License 2.0
106 stars 25 forks source link

Training has "None values not supported" #9

Closed nanometer34688 closed 4 years ago

nanometer34688 commented 4 years ago

I have tried using Tensorflow-gpu versions 1.15.0 and converted your training script to use Tensorflow-gpu 2.10.0.

My data structure is as follows:

/data
        /TRAINING_SET
            /s2_l_bgwj2n_with_s4_s4_l_pwiv4a.npy
            /s2_l_bgwj2n_with_s4_s4_l_pwiv4a.wav
            ...
        /VALIDATION_SET
        ...
        /TEST_SET
       ...

My TFrecords get produced as expected:

/tf_records
       /TRAINING_SET
           /sample_00000.tfrecords
           /sample_00002.tfrecords
           ...
       /VALIDATION_SET
       ...
       /TEST_SET
      ...

But when it comes to training using the VL2m, "None values are not supported" Have you come across this before?

dr-pato commented 4 years ago

Hi, Can you give me more information about the error? Maybe the error log will be useful..

Giovanni

nanometer34688 commented 4 years ago

Apologies. This was my mistake.

I realised I was not correctly pointing to the TF records I had previously created.