https://freemwiki.com/index.php?action=history&feed=atom&title=Openai%2F6912b0ee-f888-8005-bad4-23801314352b 2026-04-15T06:37:02Z Revision history for this page on the wiki MediaWiki 1.43.3 https://freemwiki.com/index.php?title=Openai/6912b0ee-f888-8005-bad4-23801314352b&diff=592961&oldid=prev 2025-11-11T05:32:37Z

<p>导入ChatGPT Share对话 6912b0ee...</p> <p><b>New page</b></p><div><br /> {{DISPLAYTITLE:Motion amplification explanation}}<br /> <br /> &#039;&#039;This conversation was exported from ChatGPT&#039;&#039;&lt;ref&gt;{{cite web|title=Motion amplification explanation|url=https://chatgpt.com/share/6912b0ee-f888-8005-bad4-23801314352b|publisher=ChatGPT|access-date=2025-11-11}}&lt;/ref&gt;<br /> <br /> == gpt-5 ==<br /> === User: explain ===<br /> explain <br /> # 2. 运动幅度增强（亮度保护核心算法）<br /> if motion_amplitude &gt; 1.0:<br /> base_latent = concat_latent_image[:, :, 0:1] # 首帧<br /> gray_latent = concat_latent_image[:, :, 1:] # 灰帧<br /> <br /> diff = gray_latent - base_latent<br /> diff_mean = diff.mean(dim=(1, 3, 4), keepdim=True)<br /> diff_centered = diff - diff_mean<br /> scaled_latent = base_latent + diff_centered * motion_amplitude + diff_mean<br /> <br /> # Clamp &amp; 组合<br /> scaled_latent = torch.clamp(scaled_latent, -6, 6)<br /> concat_latent_image = torch.cat([base_latent, scaled_latent], dim=2)<br /> <br /> === Assistant: This block amplifies motion in a latent video representation while preserving brightness. ===<br /> This block amplifies motion in a latent video representation while preserving brightness. Step-by-step:<br /> # Condition if motion_amplitude &gt; 1.0: → Only applies enhancement if requested amplitude exceeds normal motion.<br /> # Frame separation ``&lt;code&gt;python base_latent = concat_latent_image[:, :, 0:1] # first frame gray_latent = concat_latent_image[:, :, 1:] # subsequent frames &lt;/code&gt;`&lt;code&gt; The latent tensor holds multiple frames; first frame acts as reference.<br /> # Difference extraction &lt;/code&gt;`&lt;code&gt;python diff = gray_latent - base_latent diff_mean = diff.mean(dim=(1, 3, 4), keepdim=True) diff_centered = diff - diff_mean &lt;/code&gt;`&lt;code&gt; Computes motion signal (diff) between frames, then removes its mean over spatial dimensions to isolate motion (without brightness bias).<br /> # Scaling motion &lt;/code&gt;`&lt;code&gt;python scaled_latent = base_latent + diff_centered * motion_amplitude + diff_mean &lt;/code&gt;`&lt;code&gt; Multiplies the motion component by motion_amplitude, keeping the mean brightness unchanged (+ diff_mean restores global luminance).<br /> # Clamp and recombine &lt;/code&gt;`&lt;code&gt;python scaled_latent = torch.clamp(scaled_latent, -6, 6) concat_latent_image = torch.cat([base_latent, scaled_latent], dim=2) &lt;/code&gt;`` Limits latent values to avoid over-saturation, then reassembles the frames.<br /> <br /> Effect:<br /> Motion contrast (changes between frames) is intensified, but brightness consistency (mean value) is preserved — that’s the “亮度保护核心算法” (brightness protection core algorithm).<br /> <br /> == References ==<br /> &lt;references /&gt;<br /> <br /> <br /> [[Category:ChatGPT Shares]]<br /> [[Category:openai]]</div>

Lukegao