题目

EE699-009/EE599-009 Quiz 5 - Swin

多项填空题

In patch merging, suppose an input feature map is reshaped from (B, H×W, C) to (B, H, W, C) and then partitioned into four sub-tensors x0--x3 via slicing (each taking every other row and column). What is the shape of each sub-tensor? ==> (B, [Fill in the blank], , [Fill in the blank], , [Fill in the blank], ) What is the shape of the concatenated tensor after the line "torch.cat ([x0, x1, x2, x3], -1)"? ==> (B, [Fill in the blank], , [Fill in the blank], , [Fill in the blank], )

查看解析

标准答案

Please login to view

思路分析

The problem describes a patch merging scenario where an input feature map with shape (B, H×W, C) is first reshaped to (B, H, W, C), and then partitioned into four sub-tensors x0, x1, x2, x3 by taking every other row and column. To determine the shape of each sub-tensor: - When selecting every oth......Login to view full explanation