search

zhaipro / keras-wdsr

超分辨率
MIT License
0 stars 1 forks source link

秒记 #1

Open zhaipro opened 5 years ago

zhaipro commented 5 years ago

https://drive.google.com/drive/folders/1SRq4CLb-iMve7n32Auln03jubCSA9fNq?usp=sharing

zhaipro commented 5 years ago

https://github.com/JiahuiYu/wdsr_ntire2018 https://github.com/krasserm/super-resolution https://github.com/wmylxmj/Anime-Super-Resolution https://zh.wikipedia.org/wiki/峰值信噪比 https://zh.wikipedia.org/wiki/对数 PixelShuffler K.shape vs K.int_shape https://keras.io/zh/ WDSR(NTIRE2018超分辨率冠军)【深度解析】 https://keras.io/zh/layers/writing-your-own-keras-layers/ https://www.kancloud.cn/aollo/aolloopencv/269599

zhaipro commented 5 years ago

还看不懂的有… http://lipixun.me/2018/07/08/weight-normalization https://github.com/ychfan/tf_estimator_barebone/blob/master/common/layers.py 我总觉得这里面的写错了，啊，搞不懂。

https://github.com/krasserm/weightnorm/blob/master/keras_2/weightnorm.py 为什么要在损失函数里实现这个小功能呢？ 方便为所有卷积层添加权重归一化？ 可是损失函数在不断的进化，我们这样岂不是如同下车吗？

https://github.com/tensorflow/addons/blob/master/tensorflow_addons/layers/wrappers.py#L24 试试参考官方教程？

https://arxiv.org/pdf/1602.07868.pdf https://github.com/keras-team/keras/blob/master/keras/optimizers.py#L436 https://github.com/seanpmorgan/tf-weightnorm/blob/master/normalization.py

zhaipro commented 5 years ago

已知问题 是我不会用cv2的resize吗？ 需要增强下采样

zhaipro commented 5 years ago

提到日程，高清化摄像头，从而美化风格变换。 摄像头：640 480 大屏：1920 1080