gmayday1997 / scenechangedet Goto Github PK

@gmayday1997 I am reading your paper scene change detection, however, I can't find figure 3 (CosimNet (shown in Figure 3)) in your paper, can you send your whole paper for me? Thanks.

I've noticed strange values for loss when training and that I am unable to get the network to train.
I have been trying to use the CD2014 dataset, More specifically the PTZ/twoPositionPTZCam/ images.

I'm not sure what really to say or what documentation to provide, so if I need to add anything please let me know.
It's quite possible that I'm just doing something or many things wrong but I'd appreciate any help.

Hi, I am very interested about your paper, but I can't get the VL-CMU-CD dataset from the url you provide. Is there any other way to get the VL-CMU-CD dataset?

Hello,

I'm attempting to get your source working; however, I'm unable to locate a copy of the "deeplab_v2_voc12.pth" pretrained model file. I've looked on the DeepLab V2 models page (http://liangchiehchen.com/projects/DeepLab_Models.html), but I don't see a model trained on VOC. Do you have a link to the model file used in your code?

Thanks!

您好，能不能详细的介绍下如何训练数据，我刚接触这方面。万分感谢

I have a question regarding the layer-balancing weights β for layer 5, 6, and 7. Do you use the THRESHS = [0.1,0.3,0.5] in the cfg files for it? If so, it seems that you are never using it in the code. Does it mean you scale the loss of layer 5 by 0.1 and layer 6 by 0.3 and layer 7 by 0.5? Can you please elaborate on that.

Thanks.

I use the default batchsize=1 and program to run, but once I change it to 4, I get the following error:

***/layer/function.py", line 33, in forward
return self.scale * x * x.pow(2).sum(dim).clamp(min=1e-12).rsqrt().expand_as(x)
RuntimeError: The expanded size of the tensor (512) must match the existing size (4) at non-singleton dimension 1. Target sizes: [4, 512, 51, 51]. Tensor sizes: [4, 51, 51]

i don't know the reason

I'm interested in your paper when I see your updated articles. I'd like to run your code to see the effect. It would be great if you could send me a complete project. My email address is: [email protected]

@gmayday1997 你好，我在使用你提供的百度云下载的cd2014测试集时，发现PTZ/twoPositionPTZcam/文件夹下所有的gt_binary中的图片内容都是0. 这样完全没办法训练吧？

@gmayday1997 您好，我在訓練中發現訓練特別慢，耗時最多的就是在eval階段，請問這樣正常嗎？還有可以將你在CDNet2014上訓練好的model share給我嗎？謝謝。

I have successfully trained the model, but I don't know how to test the model, although there is a test program on this website (#19), but there is a "KeyError: 'conv1.0.weight'" error. Error message

Hi,
I want to test the model but i can't see any test script??
How have you tested the model?

你好，我在运行代码时出现如下错误，找不到该文件，请问下能提供该文件（trainval.txt）吗？

错误如下：
OSError: /media/admin228/0007A0C30005763A/datasets/dataset_/TSUNAMI\trainval.txt not found.

Hey, nice work! I have a question about loss calculation. In your training code, you resized label images to compute loss, but I think result maps should be upsampling to calculate loss or metrics. Am I right? Look forward to your reply.

Hi, when I run the code with TSUNAMI dataset. There is error of upsample.

I am currently trying to train the model on the CD2014 dataset (deeplabv2 as a backbone).
I tried two different methods for training:

• Method 1: I used the predefined parameter for learning, except for the learning rate which I set to 1e-10 (because in an earlier test, I had the impression that 1e-7 was too large). I also adjusted the code such that I can train with batch_size > 1.
• Method 2: basically the same as for method 1, but I applied gradient clipping with max_norm=5.

The problem is that the model does not converge for either training method. Did anybody also have that issue? How did you solve it? What parameters did you use for training? How long did you have to train?

So I got the pre-trained model. But how to run it? Is there a sample code to do inference with the pre-trained model?
I just want to pass two images and show the difference.

hello，I try to train and test on VL_CMU-CD dataset. while i get the f1sore 0.658, a bit lower then paper mentioned 0.71 on test set. But i use the trained model on VL_CMU-CD dataset which you provided to test and get the flscore 0.798. I am very confused,can you describe in detail the training process and parameters of the trained model you provide?

Look forward to your reply and guidance！
2021.08.26

i trained in pytorch 1.1, and get the loss as showed below

where can i change in the code to get the right result?

I downloaded cmu dataset and reproduced the experiments

But some error has occurred in the upsampling step.

Do you know the reason?

documentation of nn.Upsample for details.
"See the documentation of nn.Upsample for details.".format(mode))
Traceback (most recent call last):
File "train.py", line 282, in
main()
File "train.py", line 246, in main
label_rz_conv5 = Variable(util.resize_label(label.data.cpu().numpy(),size=out_conv5_t0.data.cpu().numpy().shape[2:]).cuda())
File "/home/nhkim/Desktop/SceneChangeDet/src/utils/utils.py", line 228, in resize_label
label_resized[:,:,:,:] = interp(labelVar).data.numpy()
File "/home/nhkim/anaconda3/envs/cosimNet/lib/python2.7/site-packages/torch/nn/modules/module.py", line 547, in call
result = self.forward(*input, **kwargs)
File "/home/nhkim/anaconda3/envs/cosimNet/lib/python2.7/site-packages/torch/nn/modules/upsampling.py", line 131, in forward
return F.interpolate(input, self.size, self.scale_factor, self.mode, self.align_corners)
File "/home/nhkim/anaconda3/envs/cosimNet/lib/python2.7/site-packages/torch/nn/functional.py", line 2509, in interpolate
raise NotImplementedError("Got 5D input, but bilinear mode needs 4D input")
NotImplementedError: Got 5D input, but bilinear mode needs 4D input

非常感谢

the webpage https://ghsi.github.io/proj/RSS2016.html. doesn't work
and I found another link http://3dvis.ri.cmu.edu/data-sets/localization/ also doesn't work.

anyone know this or you have downloaded version could share ?

thanks
Philip

I am very interested in your paper. I want to run your code to see the effect. If you can send me a complete project, it would be great. My email address is: [email protected]