You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
------------------ 原始邮件 ------------------
发件人: "Sense-X/MixMIM" ***@***.***>;
发送时间: 2023年6月14日(星期三) 凌晨1:04
***@***.***>;
***@***.******@***.***>;
主题: Re: [Sense-X/MixMIM] 关于混合损失的问题 (Issue #13)
冒昧回答,事实上在这里 unmix_x_rec ,由于 mask_ratio 为 50%,因此x1_rec和x2_rec是“互补”的。这两份“互补”的tokens再加起来,就是一张“完全”图像的reconstruction结果,对应的target也就是单张图像。
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
我在代码里看见您在计算损失函数时,只计算重建的两张图像被mask区域的损失,对于这一点我没有异议。但是在target的设置上,我看您是直接是用源单张图像作为目标图像标签,而非源混合图像作为目标图像标签。不知道我描述得是否正确,还请您解惑
The text was updated successfully, but these errors were encountered: