摘要: SSIM 公式含义: \[ SSIM(x,y)=[l(x,y)]^{\alpha}[c(x,y)]^{\beta}[s(x,y)]^{\gamma} \] 亮度: \[ l(x,y) = \frac{2\mu_x\mu_y+c_1}{\mu_x^2 + \mu_y^2 + c_1} \] 对比度: 阅读全文
posted @ 2020-11-13 14:47 wioponsen 阅读(398) 评论(0) 推荐(0) 编辑
摘要: 代码 def data_iterator(tfrecords, batch_size=2, shuffle=True, train=True, num_parallel_reads=3): # 声明TFRecordDataset dataset = tf.data.TFRecordDataset(t 阅读全文
posted @ 2020-11-13 14:36 wioponsen 阅读(630) 评论(0) 推荐(0) 编辑
摘要: ref:https://blog.csdn.net/u010454261/article/details/71268325 问题描述: error while loading shared libraries: libcudart.so.8.0: cannot open shared object 阅读全文
posted @ 2020-10-29 11:08 wioponsen 阅读(677) 评论(0) 推荐(0) 编辑
摘要: ref: https://blog.csdn.net/weixin_43593330/article/details/108491755 在设置optimizer时, 只需要参数分为两个部分, 并分别给定不同的学习率lr。 base_params = list(map(id, net.backbon 阅读全文
posted @ 2020-10-27 15:34 wioponsen 阅读(3058) 评论(0) 推荐(1) 编辑
摘要: ref: https://nervanasystems.github.io/distiller/knowledge_distillation.html (For details on how to train a model with knowledge distillation in Distil 阅读全文
posted @ 2020-10-26 09:41 wioponsen 阅读(293) 评论(0) 推荐(0) 编辑
摘要: ref:https://blog.csdn.net/g11d111/article/details/90417308 看别人的pytorch代码中,经常有些地方会有使用.contiguous(),而有些地方去掉会报错,所以就这个问题搜索了下。 该操作保证tensor内存行优先进行连续排列。某些Ten 阅读全文
posted @ 2020-10-21 11:07 wioponsen 阅读(554) 评论(0) 推荐(0) 编辑
摘要: ref: https://blog.csdn.net/weixin_38740463/article/details/89924345 最主要的特点(优点)是使用 channel shuffle 改变数据流向,大幅度减小模型参数量和计算量。 channel shuffle解决什么问题? 解决利用Gr 阅读全文
posted @ 2020-10-16 11:18 wioponsen 阅读(327) 评论(0) 推荐(0) 编辑
摘要: ref: https://blog.csdn.net/qq_33590958/article/details/104859909 BN:Batch Norm Batch Normalization (BN) is a milestone technique in the development of 阅读全文
posted @ 2020-10-15 17:05 wioponsen 阅读(604) 评论(0) 推荐(0) 编辑
摘要: ref: https://cloud.tencent.com/developer/article/1459535 除去上述链接中提到的loss,还有 SSIM (code) Contextual loss, CX (code) Contextual bilateral loss, CoBi 阅读全文
posted @ 2020-10-10 13:44 wioponsen 阅读(430) 评论(0) 推荐(0) 编辑
摘要: ref:https://www.jianshu.com/p/1826c8965499 现象: 解决: 阅读全文
posted @ 2020-10-10 10:39 wioponsen 阅读(318) 评论(0) 推荐(0) 编辑