HZIH
HZIH
> > 你好,我想知道如何预处理原始数据并将其转换为norm_u_v_os。 您能给我一些建议或预处理代码吗? > > 能给个qq或者微信请教一下这篇论文的数据预处理的一些问题吗? 我已经不做这个啦哈哈哈,但是我可以给你我的code
``` import cv2 import numpy as np def tvl1_ofcalc(path1, path2): img1 = cv2.cvtColor(cv2.imread(path1), cv2.COLOR_BGR2GRAY) img2 = cv2.cvtColor(cv2.imread(path2), cv2.COLOR_BGR2GRAY) flow = cv2.optflow.DualTVL1OpticalFlow_create() of = flow.calc(img1, img2, None) return of def minmax_norm(x):...
> 感谢感谢,就是我有个疑问,他的模型是要求输入,水平,光流,和光应变,但是作者给的数据集是一个一个光流图,直接就是把每一张光流图输入进去吗?然后通过代码算法提取光流图的水平 ,垂直,和光应变进行后续处理吗? > […](#) > ------------------ 原始邮件 ------------------ 发件人: "christy1206/STSTNet" ***@***.***>; 发送时间: 2023年10月20日(星期五) 下午3:10 ***@***.***>; ***@***.******@***.***>; 主题: Re: [christy1206/STSTNet] data preprocessing (Issue #8) import cv2 import numpy as np def tvl1_ofcalc(path1, path2): img1...
> > I am also working on it, and as part of my research, I needed to implement it on SDXL. > > ``` > > from __future__ import annotations...
I've tried to train sdxl lora with 15 images with sd-scripts, however, the result is so bad, any advice?
Thx, I've tried the method in this video but the result is bad. And when I integrate the trained checkpoint into OMG, the generated image is not satisfactory and even...
Here are some examples: 