search

xavysp / TEED

TEED: Tiny and Efficient Edge Detector
MIT License
178 stars 17 forks source link

import dataset error #3

Closed sixgod696 closed 11 months ago

sixgod696 commented 1 year ago

When reading a BIPED dataset in JSON format, an error is reported, as shown below

image
sixgod696 commented 1 year ago

I modified the database file and the import was successful, but I'm not sure which datasets B2, B3, etc. in the datasets.py refer to? The BIPED dataset I downloaded does not contain these, and I hope the author can answer them

image
xavysp commented 1 year ago

When reading a BIPED dataset in JSON format, an error is reported, as shown below

image

Hi, sorry for my late response. It seems that your json file is empty. I have uploaded my list of training to "TEED/utils/train_pair0.lst", tray opening this file.

Cheers.

xavysp commented 1 year ago

I modified the database file and the import was successful, but I'm not sure which datasets B2, B3, etc. in the datasets.py refer to? The BIPED dataset I downloaded does not contain these, and I hope the author can answer them image

You just need "BIPED", the others come from DexiNed modules. Do not take into account the other versions.

Have a good day.

sixgod696 commented 1 year ago

When reading a BIPED dataset in JSON format, an error is reported, as shown below

image

Hi, sorry for my late response. It seems that your json file is empty. I have uploaded my list of training to "TEED/utils/train_pair0.lst", tray opening this file.

Cheers. Tnank you for your answer. I have expanded the BIPED data set and completed the training. Now I have a problem with the evaluation. The script I found on the Internet to calculate the ODS and OIS used by hed is only 0.06. I am not sure what is causing this? Is it an issue with the thickness of the line?

xavysp commented 1 year ago

Hi, as maybe you can see, results from TEED are thinner than the ones from HED, even DexiNed. So I am preparing the Dollar's code for TEED. In the meantime, try the evaluation used in DexiNed, with the following modification:

tmp_edge = edgesNmsMex(tmp_edge,O,1,5,1.03,8);
sixgod696 commented 1 year ago

Hi, as maybe you can see, results from TEED are thinner than the ones from HED, even DexiNed. So I am preparing the Dollar's code for TEED. In the meantime, try the evaluation used in DexiNed, with the following modification:

tmp_edge = edgesNmsMex(tmp_edge,O,1,5,1.03,8);
image image

I used the evaluation code of the hed integration, I changed your parameters, the image processed by the model was preprocessed to look like this, and the evaluation result remained unchanged. I'm new to MATLAB, and I'm not sure how to use and modify the evaluation code you provided.

xavysp commented 1 year ago

Hi, sorry for my late response, I am changing the city, so these days are chaotics. So you converted GT images in MAT? I do not remember the BSDS GT mats, but they are binary, (0,1) did you checked that?

sixgod696 commented 1 year ago

Hi, sorry for my late response, I am changing the city, so these days are chaotics. So you converted GT images in MAT? I do not remember the BSDS GT mats, but they are binary, (0,1) did you checked that?

Thank you for your reply, I have solved this issue due to my lack of understanding of the edge inspection inspection process, and the test results have been successfully displayed so far.

xavysp commented 1 year ago

Oh great, glad you are using TEED.