I've been trying to use tensorlayer.prepro.threading_data, but I'm getting a different return type for different inputs. Sometimes it returns an ndarray and sometimes it returns a list. The documentation doesn't specify what's the reason for the different return types.
Can anyone shed some light on this?
Answer: It seems that what was causing this problem was having items with different shapes in the list. In this instance, PNG images with 3 and 4 channel. Removing the alpha channel (the fourth channel) from all PNG images solved this for me.
It seems that what was causing this problem was having items with different shapes in the list. In this instance, PNG images with 3 and 4 channel. Removing the alpha channel (the fourth channel) from all PNG images solved this for me.