issues Search Results · repo:Fediory/HVI-CIDNet language:Python
Filter by
70 results
(73 ms)70 results
inFediory/HVI-CIDNet (press backspace or delete to remove)非常感谢您的工作，我近期研读了您的论文，并尝试将他应用到raw
construction的工作中，但是预测出来的结果与gt色调上会有一定的差别，再次研读了您的论文，想尝试将gammaG和gammaB作为超参数送到网络中学习，你认为这种方式能否起到帮助呢？
smy123666
- 5
- Opened 2 days ago
- #74
faxiangui作者您好，读了您的这篇工作和您在问题集处的解答，我收获很大。您针对低光照图像增强领域的切入点，让我敬佩。但是，作为一个新手，我该怎么去找到研究领域内存在的问题？比如您发现HSV色彩空间的不足一样？最初是通过什么途径去发现存在这种问题的？希望能够得到您的回复，感激不尽！
Wuko-dev
- 1
- Opened 3 days ago
- #72
请问在4060Ti显卡 14核CPU上，batch size=8，epoch=1500，thread=14，训练lolv1数据集，一个epoch要花费26分钟，正常吗
lxr1137805288
- 5
- Opened 5 days ago
- #70
igodogiHello,
First of all, thank you for your work on HVI-CIDNet! I have used this model for low-light enhancement, which seems to be
its primary purpose. The results have been impressive.
I wanted to ask ...
danishbansal808
- 1
- Opened 7 days ago
- #68
作者您好，这是我第一篇复现的work，有几个问题想向您请教，烦请拨冗。 1，我连续进行了两次实验，都是在LOLv1数据集上进行，第一次是实验是按照您在 #16 中提及的超参数训练，最好的结果是| 280 | 22.3457 | 0.8092
| 0.1905 |，第二次实验是按照您源代码中的参数原封未动进行试验，最好结果是| 270 | 23.0486 | 0.8103 | 0.1540 |，想请问没有达到原文中的指标可能是什么原因？ ...
Cloudyu1215
- 8
- Opened 9 days ago
- #67
作者您好，我使用的RTX4090(24G), Python 3.10.14， torch = 2.2.2+cu12,and torchvision=0.17.2,实验设置的超参全部与分享的代码相同，但是指标只有：PSNR=23.5881 ,
SSIM= 0.8539,请问可能是什么原因导致结果偏低？
- 4
- Opened 10 days ago
- #66
作者大大论文里面写的是batchsize为8，cropSize为400，我在4090上面试了一下，一块不太行，需要两块才能跑，作者大大知道多GPU该咋改吗，看您代码中是用单块GPU跑的
3d11123
- 2
- Opened 13 days ago
- #65
Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip!
Press the /
key to activate the search input again and adjust your query.Learn how you can use GitHub Issues to plan and track your work.
Save views for sprints, backlogs, teams, or releases. Rank, sort, and filter issues to suit the occasion. The possibilities are endless.Learn more about GitHub IssuesProTip!
Restrict your search to the title by using the in:title qualifier.