Getting an error while using I3D to train baseline classifier on SSBD

[kevinium]

I am trying to train the baseline clarifier using the SSBD dataset. I have clipped the SSBD dataset to include only the parts that show stimming behaviours, I have extracted the flow using Preprocess_threads.py and made batches of 8 using utils.py

However when I use train_flow.py to train the baseline model I encounter the following error

ValueError: Shapes (None, None) and (None, None, 1, 1, 1024) are incompatible

When I modify the code to not reshape x

def generate_arrays_from_file(data, labels):
        while True:
                for i in range(len(labels)):
                        #x, y = data[i], labels[i]
                        x, y = np.load(data[i]), labels[i]
                        #x = x.reshape(1, x.shape[0], x.shape[1], x.shape[2], x.shape[3])
                        print(x.shape)
                        yield x, y

I get the following error

ValueError: Exception encountered when calling layer "i3d_inception" (type Functional).
    Input 0 of layer "Conv3d_1a_7x7_conv" is incompatible with the layer: expected min_ndim=5, found ndim=4. Full shape received: (None, None, None, None)   
    Call arguments received:
      • inputs=tf.Tensor(shape=(None, None, None, None), dtype=float32)
      • training=True
      • mask=None