I use mxnet on insightface to extract features.
The output of the model is a feature, but insightface sets the label for forward().
db = mx.io.DataBatch(data=(_data,), label=(_label,))
and did not enter a label when loading the model.
model.bind(data_shapes=[(‘data’, (args.batch_size, 3, image_size, image_size))])
I don’t think you should enter a label in the feature extraction to avoid errors.