Created by: HaiLongEmb

配置信息如下： metric: VOC pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/yolov3_r50vd_dcn_obj365_dropblock_iouloss.tar #pretrain_weights: https://paddlemodels.bj.bcebos.com/object_detection/ResNet50_vd_dcn_db_obj365_pretrained.tar weights: output_test_iouloss/yolov3_r50vd_dcn_db_iouloss_obj365_pretrained/model_final num_classes: 3 finetune_exclude_pretrained_params: ['yolo_output'] use_fine_grained_loss: True

但运行提示:

Error Message Summary:

Error: The input's size along the split dimension must be evenly divisible by Attr(num_or_sections). But received Attr(num_or_sections) = 8, input(X)'s shape = [4, 50, 4], Attr(dim) = 0. [Hint: Expected input_axis_dim % num == 0, but received input_axis_dim % num:4 != 0:0.] at (D:\1.7.0\paddle\paddle/fluid/operators/split_op.h:41) [operator < split > error]

为什么输入的尺寸会有区别？

这个问题我找到了，YOLOv3Loss中的batch_size设置没有与TrainReader.batch_size一致