為啥切臉速度這麼慢？

main9dog

星级打分

• 1
• 2
• 3
• 4
• 5

平均分:NAN  参与人数:0  我的评分:未评

之前用1060 後來用4070，感覺每秒速度才快1張不到 ？


版上看有兩個版本

1.免費的佛心作者做的LOGO就一隻貓貓（但聽說這個眼睛位置偶爾會切歪？）

2.付費的（ 台灣沒再用QQ 想付費也沒辦法付費 ）

到現在沒有其他替代方案嗎？

所以到底為啥原版切臉速度這麼慢？ 是因為切的最準嗎？

seancai110

dfl识别用的是gpu。但是图像处理用cv2的时候，用的是cpu

Cxsmo

Quick face cut is my preferred one to use

maymay

Cxsmo 发表于 2025-3-30 07:46
Quick face cut is my preferred one to use

In fact, the original DFL cutting speed is not slow and has high precision.
The problem with the brother above is that he is using a 4070 graphics card and is using the DFL30 series version. The DFL30 series version of TensorFlow is not natively supported because the 30 series version of TensorFlow is not compatible with the 40 series graphics card, so the face cutting speed will be slower than the previous version, which has no impact on training.

Cxsmo

maymay 发表于 2025-3-30 11:50
In fact, the original DFL cutting speed is not slow and has high precision.
The problem with the b ...

No dfl has lower speed and lower accuracy compared to quick face cut which is a good amount better

main9dog

Cxsmo 发表于 2025-3-30 13:42
No dfl has lower speed and lower accuracy compared to quick face cut which is a good amount better

所以版上那個

快速切脸V2.0.12

實際就是原版切臉好？

Cxsmo

main9dog 发表于 2025-3-30 21:23
所以版上那個

快速切脸V2.0.12

.that version is the latest quick face cut  

DFLive

maymay 发表于 2025-3-30 11:50
In fact, the original DFL cutting speed is not slow and has high precision.
The problem with the b ...

原版还不慢？那快切和魔切是闲得没事干写出来的吗。

main9dog

快切還是輸原本切......

最近嘗試一部，稍微有一點往右邊側臉（五官都還是看的到喔）
右眼跟臉融合出來就怪怪的.....那幾幀用正常切就正常..... 我猜魔切應該也是吧＝＝