Zhuoran Zheng
Zhuoran Zheng
您好，我们会上传完整的pipline，请等候。 ---------- 该邮件从移动设备发送 --------------原始邮件-------------- 发件人："wuli超超 ***@***.***>; 发送时间：2023年3月30日(星期四) 下午4:15 收件人："zzr-idam/4KDehazing" ***@***.***>; 抄送："Subscribed ***@***.***>; 主题：[zzr-idam/4KDehazing] 数据集生成 (Issue #15) ----------------------------------- 请问文章中的生成雾图的代码会上传么，如与真实雾图域适应的损失等 — Reply to this email directly, view it on GitHub, or unsubscribe....
In the paper, we use the l2 loss function as targer, but effect of the l1 is good.
Sorry, this project has been optimized for many times. I suggest you can train again on it, and the effect will not be much worse. ------------------ Original ------------------ From: ***@***.***>;...
We have added 4000 new pairs of images to the original dataset. You can train a better model from the latest dataset.
Our model is quick for inference and may have something to do with your cold start.
Sorry, 1) our model has been improved to improve the performance, such as using Unet instead of convolutional blocks; 2) the pre-trained model we will retrain.
ok
We used the evaluation code provided by kornia.