During inference on synapse detection, we will be loading the pre-trained model. The sample_label_size is not initialized globally leading to attribute errors.
Thanks for finding the bug! The problem is defining label_vol_ratio when label is None. Therefore I put label_vol_ratio definition under 'if self.label is not None: '.
During inference on synapse detection, we will be loading the pre-trained model. The sample_label_size is not initialized globally leading to attribute errors.