Question

我用keras mnist数据集训练了一个用于手写数字识别的模型，它的准确性为98％。但是，就我自己的形象而言，性能却很差。我想这与我自己的图像的预处理有关。这是我尝试将图像转换为28 * 28尺寸的方法。

    ; with test as
    (
        select  *, 
                rn         = row_number() over (order by Created desc),
                approv_rn  = row_number() over (partition by EmailApproved 
                                                    order by Created desc)
        from    @Test
    )
    select  *
    from    test t
            outer apply
            (
                select  x.rn
                from    test x
                where   x.EmailApproved = 1
                and     x.approv_rn     = 2
            ) x
    where   t.rn    < x.rn or x.rn is null
    order by t.Created desc

我发现转换后图像的颜色发生了变化

here's the original image

image after resize

您可以看到变换后白色变成灰色，我想知道这是否是性能不好的原因吗？

Answer 1

尝试使用opencv读取图像。

 ResponseEntity<String> response = restTemplate.exchange(restApiUrl,
                HttpMethod.POST, entity, String.class);
        LOGGER.info("XML Response from UIDAI :" + response.getBody()); //** Error**//

用MNIST数据集训练的模型预测自己的图像

1 个答案: