zzr-idam / 4kdehazing Goto Github PK

Could you give me a Google Drive link to the daytime 4K dataset?

Thank you for you reply.

Sorry, excuse me, the data set link is out of date, could you republish it?

Hello, thanks for sharing your dehazing work.

It seems that the shared pretrained model does not exactly match the network structure described in the paper. I have also conducted some tests and the inference quality of the shared pretrained model is not quite desirable when applied for 4KID or O-Haze dataset.

So could you please briefly explain how the shared pre-trained model is generated? Is it for 4KID or for O-Haze dataset? Are there any fine-tunes needed for models used for O-Haze dataset?

Thanks!

I used 3 Quadro RTX 6000 (24GB) GPU cards for training. Why can I only train over 100 Epochs a day ？
May I ask what is the reason for this? Has anyone had a similar situation?

您好！用你的模型进行你提供的数据集图像进行测试，去雾后的图像怎么还变模糊了呢。想请教下

是有模型优化还是其他的推理加速方式呢？

Hi, thanks for your contribution.
when I read your paper, you use l2 loss as your loss function.
However, in your code, it seems like you used l1 loss, am I right?

Train.py：(line 25)
mse = nn.L1Loss().cuda()

so, the result in your paper is obtained by L1 or L2 loss?

你好，请问深度图能公开一下吗，谢谢！

In your work（Single UHD Image Dehazing via Interpretable Pyramid Network）, do you have the 4KID dataset link? Thank you for your work.

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

Hi. Great work and improvement. I would really like to ask you about 2 things.

1. Can this approach be used with hd or full images?
2. Could you please share the pretrained model or dataset. I cannot access to the baidu drive. It had a problem. Could you please share it? It could be great to test it. Thanks and congrats

Thanks for your contribution.

But the inference time I calculated is quite different from it mentioned in the paper. It takes more than 100ms for a 4K image. The code I use is as follows.
'''
model = B_transformer().cuda()
a = torch.randn(1, 3, 1024, 1024).cuda()
starter, ender = torch.cuda.Event(enable_timing=True), torch.cuda.Event(enable_timing=True)
repetitions = 100
timings = np.zeros((repetitions, 1))
# GPU-WARM-UP
for _ in range(50):
enhanced_image = model(a)
# MEASURE PERFORMANCE
with torch.no_grad():
for rep in range(repetitions):
torch.cuda.synchronize()
starter.record()
enhanced_image = model(a)
ender.record()
# WAIT FOR GPU SYNC
torch.cuda.synchronize()
curr_time = starter.elapsed_time(ender)
timings[rep] = curr_time
mean_syn = np.sum(timings) / repetitions
std_syn = np.std(timings)
print(mean_syn)
'''
Is this right?

In your latest work, Single UHD Image Dehazing via Interpretable Pyramid Network.
In the paper, it is mentioned that the noise is suppressed by using Tucker, but there is no code related to Tucker in the code you provided. May I ask in which step you use Tucker
Thank you for your work and look forward to your reply

Learned another article from your team：Ultra-High-Definition Image HDR Reconstruction via Collaborative Bilateral Learning。Can I learn the code of the loss function in this article from your team：Lse。
thanks a lot.

可以看看你的文献完整模型测试代码吗？关于《基于可解释金字塔网络的单张超高清图像去雾》这篇文献的
谢谢啊

Thank you for your impressive work!.

If you don't mind, can you share pre-trained model?

Thank you!

Hey,
the dataset links are out of date, can you please update that

Thank you for your code. Can the project run in Windows system？Please,thanks

Could you please provide a Google Drive or Dropbox link of your dataset?