benchmark icon indicating copy to clipboard operation
benchmark copied to clipboard

SEResnet50 模型定义问题

Open imluoxu opened this issue 6 years ago • 1 comments

在CV的benchmark上,SE-ResNeXt50模型的定义地址为 https://github.com/PaddlePaddle/models/blob/develop/PaddleCV/image_classification/models/se_resnext.py的文件。有以下代码: if layers == 50: cardinality = 32 reduction_ratio = 16 depth = [3, 4, 6, 3] num_filters = [128, 256, 512, 1024]

        conv = self.conv_bn_layer(
            input=input,
            num_filters=64,
            filter_size=7,
            stride=2,
            act='relu',
            name='conv1', )
        conv = fluid.layers.pool2d(
            input=conv,
            pool_size=3,
            pool_stride=2,
            pool_padding=1,
            pool_type='max',
            use_cudnn=False)

根据这里的定义 weight 的深度结构是[3, 4, 6, 3] 个数定义为 [128, 256, 512, 1024] 标准的Resnet 50的 weights 定义应该是 [256,512,1024,2048], 是不是相当于SE-RESNET是为了降低计算的开销,所以做了简化设计?

imluoxu avatar Jul 30 '19 09:07 imluoxu

SE-ResNeXt50和Resnet 50是两个不同的模型,SE-ResNeXt50在channel定义上比Resnet 50要小

heavengate avatar Jul 30 '19 12:07 heavengate