인프런 커뮤니티 질문&답변

def inception_module(x, filters_1x1, filters_3x3_reduce, filters_3x3, filters_5x5_reduce, filters_5x5, 

                     filters_pool_proj, name=None):

    '''

    x: 입력 Tensor

    filters_1x1: 단독 1x1 필터수

    filters_3x3_reduce: 3x3 Conv 적용 전 1x1 Conv 필터수

    filters_3x3: 3x3 Conv 필터수

    filters_5x5_reduce: 5x5 Conv 적용 전 1x1 Conv 필터수

    filters_5x5: 5x5 Conv 필터수

    filters_pool_prj: MaxPooling 적용 후 1x1 Conv 필터수 

    '''

    

    # 첫번째 1x1 Conv

    conv_1x1 = Conv2D(filters_1x1, (1, 1), padding='same', activation='relu')(x)

    # 3x3 적용 전 1x1 conv -> 3x3 Conv

    conv_3x3 = Conv2D(filters_3x3_reduce, (1, 1), padding='same', activation='relu')(x)

    conv_3x3 = Conv2D(filters_3x3, (3, 3), padding='same', activation='relu')(conv_3x3)

    # 5x5 적용 전 1x1 Conv -> 3x3 Conv

    conv_5x5 = Conv2D(filters_5x5_reduce, (1, 1), padding='same', activation='relu')(x)

    conv_5x5 = Conv2D(filters_5x5, (5, 5), padding='same', activation='relu')(conv_5x5)

 

    pool_proj = MaxPooling2D((3, 3), strides=(1, 1), padding='same')(x)

    pool_proj = Conv2D(filters_pool_proj, (1, 1), padding='same', activation='relu')(pool_proj)

    # 단독 1x1 결과, 3x3 결과, 5x5 결과, pool이후 1x1 결과 feature map을 채널 기준으로 Concat 적용. 

    output = Concatenate(axis=-1, name=name)([conv_1x1, conv_3x3, conv_5x5, pool_proj])

    return output

 

빨간색 (x)부분들에  진행하지않은 X 가 아닌 1*1진행한 conv_1*1 들어가야 맞는거 아닌가요`?