How to detect the edge of our own data

[xwhkkk]

Thank you for sharing your great job !
How can we detect the edge of our own jpg or png files? Can we directly use the pre-trained model?

Thank you for your suggestions!

[MengyangPu]

Hi, you could try to use the pre-trained model to detect your images.
Firstly, prepare a list of images according to the file test.txt of BSDS. The first column is the path of your image. The second column can be set to a path ending with .mat, which is only used for placeholders.
Second, Refer to the tutorial for detection.

[BugMaker-bot]

Hi, you could try to use the pre-trained model to detect your images. Firstly, prepare a list of images according to the file test.txt of BSDS. The first column is the path of your image. The second column can be set to a path ending with .mat, which is only used for placeholders. Second, Refer to the tutorial for detection.

I saw your reply earlier that a part of the code needs to be changed when the gt is only png, but I can't find it now, can you point out which part again please?

[Arslan-Mehmood1]

Hi, you could try to use the pre-trained model to detect your images. Firstly, prepare a list of images according to the file test.txt of BSDS. The first column is the path of your image. The second column can be set to a path ending with .mat, which is only used for placeholders. Second, Refer to the tutorial for detection.

and where do we need to put the path of test.txt?
The process to make inference is not clear, in tutorial there is no guidance on setting the custom test images!
isn't there any way of using the pretrained checkpoints to make direct predicitons instead of using this long process of test.py?

[MengyangPu]

@BugMaker-bot
Currently, reading data consisting solely of PNG files is not supported. However, please rest assured that MAT files are not being used. Refer to the test.txt file to create your data list.
If you want to attempt modifying the data loading function, you can refer to here

[MengyangPu]

@Arslan-Mehmood1
Refer to the test.txt file to create your data list.

[MengyangPu]

@Arslan-Mehmood1
Firstly, you can place your data wherever you find suitable, for example, /EDTER/data.
Second, create a test.txt file based on the absolute path of your data. The first column is the path of your image. The second column can be set to a path ending with .mat.(please rest assured that MAT files are not being used.)
Finally, refer to configs/base/datasets/bsds.py to create your custom data configuration file.

[MengyangPu]

@Arslan-Mehmood1
Yes, you can use the pretrained checkpoints for direct predictions. Instead of using the test.py script, you can load the pretrained model checkpoint in the PyTorch framework and then use the model to make predictions on new data. But this may require you to customize the test script.
The current test script is modified based on mmsegmentation. You can refer to the original script file for guidance.

[Arslan-Mehmood1]

thanks @MengyangPu for such a detailed response.
I'll work on it and share my feedback

best regards,
Arslan

[2783000]

thanks @MengyangPu for such a detailed response. I'll work on it and share my feedback

best regards, Arslan

Could you share how you did it?

[7zburger]

Can you edge-eval on your own dataset?

[grzhao89]

I tested on my own dataset, stage I worked well but stage II didn't. It kept showing error that 2 tensor dimensions are different on fuse head part. I don't want to resize my images, so I got stuck there.

[Gao-yudong]

I tested on my own dataset, stage I worked well but stage II didn't. It kept showing error that 2 tensor dimensions are different on fuse head part. I don't want to resize my images, so I got stuck there.
Hi, I encountered the same issue. Have you resolved it?