Menghan Xia comments

Results 12 comments of


                                            Menghan Xia

Shape mismatch in tuple component 0. Expected [256,256,3], got [480,640,3] [[{{node input/batch/fifo_queue_enqueue}}]] upon testing dataset and also throwing this exception too.

To test on images of other resolutions, you have to modify the hard-coded image resolution variables in './model.py' (line:7~8). Btw, the image width and height have to be multiple of...

What's the difference between in_img and gt_img?

Yes, the in_img and gt_img is the same. So, only one of them is needed.

Reproduced the CartoonGan

Yes, I met the same problem. In my opinion, it is because the CartoonGAN is only feasible to those photos that have a very similar appearance to cartoon style themselves....

您好，我在训练ReversibleHalftoning的代码时，下载的逆半色调网络的预训练数据包是损坏的，请问您可以重新更新一个吗？

你好，如果你下载的ckpt是损坏的，那可能是因为你的网速太慢导致，可以考虑在更高网速环境重新下载试试。

Sample dataset

Any image sequence with overlap, ordered or unordered, could be stitched by our approach. For example, you can take some photos of flat floor using a smart phone. Alternatively, you...

您好，我对您的这个工作非常感兴趣，想了解下训练数据halftone的goundthruth是怎么生成的呢？

你好！严格说我们的模型训练没有ground-truth halftone，因为在我们工作之前不存在invertible halftone的生成方法。实际上，我们训练中用到的halftone images (error-diffusion) 只是在warm-up阶段用来稳定模型训练用的，而这些halftone images是通过传统的halftoning算法[1]从灰度图计算得到的。 [1] Victor Ostromoukhov. A simple and efficient error-diffusion algorithm. In ACM SIGGRAPH, 2001