nie-lang / deepimagestitching-1.0 Goto Github PK

============
Traceback (most recent call last):
File "inference.py", line 76, in
inference_func(pretrained_model)
File "inference.py", line 64, in inference_func
input_clip = np.expand_dims(input_loader.get_video_clips(i), axis=0)
File "D:\Project\depth\DeepImageStitching-1.0-main\Codes\Stitch_Net\utils.py", line 74, in get_video_clips
image = np_load_frame(video_info_list[i]['frame'][index], self._resize_height, self._resize_width)
IndexError: list index out of range

Excuse me, would you mind offer your pretrained model for image stitching with Baidu Cloud? I can't open it with Google Drive. Thanks a lot.

2022-06-17 20:22:14.965496: W tensorflow/core/framework/op_kernel.cc:1273] OP_REQUIRES failed at save_restore_v2_ops.cc:184 : Not found: NewRandomAccessFile failed to Create/Open: ../checkpoints/stitch/model.ckpt-600000.data-00000-of-00001 : ϵͳ�Ҳ���ָ�����ļ���; No such file or directory

Hello. You have provided the pretrained model for stitching network. But I cant find pretrained model for deep homography network. Could you provide it? Thanks!

Hi, firstly thanks for opening your source code. It is really an amazing job!
Recently, I'm trying to reproduce your code in Pytorch. And I have a question about the content revision stage, specifically in UNet. I found that the function(maxpool2d in tensorflow) you use will keep the feature map size. But it seems hard to use Pytorch to do the same. Is this necessary for the result?

(I've finished the Homography estimation part, and its coarse result is similar to yours. But the result of content revision stage is pretty strange(Even when I seperately train the unet using coarse image from yours ). So I searched for your code and found that your maxpool can keep the size while mine can't). The Unet I use and the result of Unet is as follow:

Result

Could you please give me a PDF version of the paper?
Thank you!
email:[email protected]

How to solve this bug?

Hello, thanks for your code! I modified the utils function according to your requirements under Windows OS, but it shows "list index out of range"? For Dataset_gen/ dataset.py is OK.

Looking forward to your reply.

Hi, I cannot understand the "Auxiliary matrices used to solve DLT" in the "tensorDLT.py" . Can you tell me the solution process in more detail?

In the process of preparing dataset,
Ⅰ) I created folders of training and testing, these two folders are empty of course.
Ⅱ) I created a folder named row_image in which I put some pictures (.jpg)
Ⅲ) After that, I Set the path for row images, training samples, and testing samples in ./Dataset_gen/dataset.py as you said in the github project.
I think if I run dataset.py, the program will automatically generate something in training folder and testing folder according to the stuff in folder named row_image, but the program can't run successfuly, the error is as follows: IndexError: Cannot choose from an empty sequence, could you tell me why, I am confused because my raw_image is not empty, there are some pictures(.jpg) in it.