aiueogawa

@JiahuiYu Thank you for your code and paper! I also have a question about the DeepFillv2 discriminator. Do all layers in discriminator remain conv2d instead of gated conv2d?

@JiahuiYu Thank you for your response! I also think discriminator with gated convolutions would work better for gated convolutions could pay attention to the part around the mask and user...

@JiahuiYu I'm a bit confused about the **activation function at the last layer in the discriminator** in DeepFillv2. You said two different things about it. 1. you used the same...

@JiahuiYu Thank you! I see. As you say, leaky ReLU also seems good and is. In terms of the magnitude of gradients, leaky ReLU seems better than tanh, gradients of...

@JiahuiYu What hyper parameters, maxVertex, maxLength, maxBrushWidth and maxAngle, did you use for free-form mask generation especially in CelebA-HQ? And how many strokes in a free-form mask did you use?

@JiahuiYu Thanks for your answer. `min_num_vertex`, `mean_angle`, `angle_range`, `min_width` are not in your free-form mask generation algorithm. I expect them to be as follows: - `num_vertex` is chosen from the...

@khemrajrathore Your code still involves bugs around drawing a circle at a joint point, e.g. your joints are too big, as @JiahuiYu said. The radius of a circle should be...

@JiahuiYu Thanks for revised information. Q1: Your discriminator of SN-PatchGAN produces HxWxC outputs and the loss defined in the following definition is also the same size HxWxC, because ReLU is...

@Puzer +1

dockerize、dockerizationは **「Docker化」と訳出すべきだと思います** 。 Google検索で「Docker化」と「Docker対応」のそれぞれでヒットする件数を比較してみると、前者は1,470,000件、後者は755,000件と約2倍の差があります。私の感覚からもこの結果からも「Docker対応」よりも「Docker化」の方が自然な表現でかつ意味を的確に表した表現だと思います。 ちなみに、「Docker対応」の方はWindowsがDockerに対応したというニュースに関連する記事がトップにヒットするようです。