Skip to content

v0.5.1 找地面不误判、脚掌贴地、深度图更省

Choose a tag to compare

View all tags
@extrant extrant released this 28 Oct 17:33
· 1 commit to main since this release
v0.5.1
9e7e735

更新方法

  • 插件：插件仓库自动更新到 0.5.1。
  • 后端（一键包）：设置页点「更新后端」。会照样检查面捕、深度图组件，缺的自动补齐（从 0.4.0 及更早版本升级的会下载约 64MB + 400MB；0.5.0 已经装好的不用重下）。下载不了可以看 手动更新后端。
  • 已识别过的图读的是缓存，点工具栏的「重新识别当前图片」才会用上新的找地面和深度图。

新内容

  • 找地面更准：只在看得见脚时找地面，并检查人确实站在那个面上。半身照、特写不会再把背景墙、衣服当成地面，人不会被扳歪（之前可能歪 17~68°）；俯拍等真有地面的照片照样扶正
  • 脚掌贴地：修正了脚的静止姿态对应，站着的人不再脚尖翘起、只有脚后跟着地
  • 新开关「踩地的脚放平」（调整页 → 站姿，默认开）：踩在地上的脚换成游戏里平放的样子，只转脚、不动腿；踮脚、勾脚这类明显故意的不会动
  • 深度图更省：分辨率调到 1200 块，结果基本不变（30 张照片对比，地面判断完全一致），显存峰值从约 1.4GB 降到约 0.75GB，CPU 模式从约 3 秒降到约 1 秒
  • 后端更新照样会检查并补齐面捕、深度图组件；从 0.4.0 及更早版本升级的，在设置页点「更新后端」一次装好
  • 改了识别相关的设置：已识别过的图片读的是缓存，点工具栏的「重新识别当前图片」按钮才会用上新的找地面和深度图

How to update

  • Plugin: updates to 0.5.1 automatically through the plugin repository.
  • Backend (one-click package): click "Update backend" on Settings. It still checks for the face capture and depth components and adds any that are missing (about 64 MB + 400 MB when coming from 0.4.0 or earlier; 0.5.0 users already have them). If downloads fail, see Updating the backend by hand.
  • Recognized images come from the cache: use the toolbar's "Recognize the current image again" button to apply the new floor detection and depth settings.

What's new

  • Better floor detection: the floor is only searched for when the feet are visible, and the person must actually stand on it. Half-body shots and close-ups no longer mistake a background wall or clothing for the floor and get tilted (previously by 17–68°); photos with a real floor, such as high-angle shots, are still straightened
  • Feet flat on the ground: the feet's rest pose is now matched correctly, so standing characters no longer tilt their toes up and touch the ground with the heel only
  • New "Flatten planted feet" option (Tuning → Stance, on by default): feet standing on the ground are laid flat like in game, turning only the foot and leaving the legs alone; clearly deliberate tiptoe or flexed feet are kept
  • Lighter depth: the depth model runs at 1200 tokens with practically the same results (identical floor decisions on 30 test photos); peak VRAM drops from about 1.4 GB to about 0.75 GB, CPU mode from about 3 s to about 1 s
  • Backend updates still check for and add the face capture and depth components; updating from 0.4.0 or earlier needs one click on "Update backend" in Settings
  • Recognized images come from the cache: use the toolbar's "Recognize the current image again" button to apply the new floor detection and depth settings
Assets 3
Loading