DALLEEncoder

class mmpretrain.models.selfsup.DALLEEncoder(group_count=4, n_hid=256, n_blk_per_group=2, input_channels=3, vocab_size=8192, device=device(type='cpu'), requires_grad=False, init_cfg=None)

DALL-E Encoder for feature extraction.

参数:

• group_count (int) – Number of groups in DALL-E encoder. Defaults to 4.

• n_hid (int) – Dimension of hidden layers. Defaults to 256.

• n_blk_per_group (int) – Number of blocks per group. Defaults to 2.

• input_channels – (int): The channels of input images. Defaults to 3.

• vocab_size (int) – Vocabulary size, indicating the number of classes. Defaults to 8192.

• device (torch.device) – Device of parameters. Defaults to torch.device('cpu').

• requires_grad (bool) – Require gradient or not. Defaults to False.

• init_cfg (Union[List[dict], dict], optional) – Config dict for weight initialization. Defaults to None.

forward(x)

Forward function of DALL-E encoder.

参数:

x (torch.Tensor) – The input images with shape (B, C, H, W).

返回:

The output with shape (B, vocab_size, h, w).

返回类型:

torch.Tensor