한국어 condition 변수 만드는 법

[internationalwe]

Condition consists of labels, strokes, style.

Labels are created with a size of (batch, 100) through embedding,
Strokes are counted based on their frequencies for initial, medial, and final strokes.

For the style, after the style encoder, the resulting vector has a dimension of 32,768 when flattened. Therefore, there is a need to compress the output into a smaller size, and three methods have been tried:

1. Using nn.linear to reduce it to a size of (batch, 128).
2. Utilizing nn.conv2d, maintaining the channel dimensions while setting the kernel size to be the same as the height and width, performing convolution operations to obtain a size of (batch, 128, 1, 1) and flattened to (batch, 128).
3. Using adaptive_avg_pool2d for global average pooling, resulting in a size of (batch, 128, 1, 1) and flattened to (batch, 128)