captanlevi / contour-detection-pytorch Goto Github PK

Hi @captanlevi

I have encountered one more doubt while trying to use my own dataset.

From the mentioned code I understand that the targets to the model are the segmented images not the improved_contours (extracted from the given script). In this case how does the model learns to predict specific boundary for a given object? And where exactly these extracted contour images (improved_contours) are used. Because I don't see anywhere its used.

Thanks again

Hi @captanlevi

Thanks again. I was successfully able to start the training. However, the loss value becomes constant after 5 iterations. Could you please suggest me some possible solutions that I can try?

Thanks in advance

Hi @captanlevi

I had the issue with the multiprocessing as mentions in this. And I followed your suggestion to create a dataloader without mutliprocessing. But I end with the mentioned error in below line.

I am using (224, 224) as image size. Could you please suggest me what I should be checking when I have this error?
Thanks in advance

Hi, the link of the best model can't be opened, can you send the model to my email [email protected]

Hi thank you for your help

so I solve the path problem

I try to train with bdd data, but after completing the training and testing, no edge was detected

here is bdd segmentation data and edge data

Hi again,
I could train successfully, but for evaluation step, you mentioned that there is make_results.py which should be run. But, there is no such python file in your repository. Is there any?

Thank you!

Hi thank you for your help

because of you, I train my own model successfully

I want to evaluate models but there is no "make_result.py" so can I get that?

Hi @captanlevi

Can you please provide make_results.py file for evaluation?

Thanks

hi thank you for your fantastic code

I want to extract the result file for each image by putting a folder with multiple image files as input

I want to get the extracted data of the contour

can you help me?

hi your code is very useful for me! thank you for your kind and smart

I wanted to train the autonomous vehicle data, so I tried to train 7000 pieces of data, but first I trained only 181 pieces of data to match the name with train.txt.
but I got some problem KeyError: 'model_save_path' error can I help you?

Hello, thank you for your great code.

I have some problem about 'Q' which loads data.
It can't put data (Q.put doesn't work), so in start_training, shared_Q is always empty.
I don't know why it doesn't work even though I think both image and text files exist.

Have you ever had a problem like this or can you help me?
Thank you again :)

Thank you for sharing your work!
I am interested in applying this model. In Readme it is said: Download the DenseCRF refined segmantation maps for PASCAL-2012 dataset. And place them in SegmentationObjectFilledDenseCRF folder.

Could you please let me know where can I find it?

Hi @captanlevi

Thanks for the great work. I can access the segmentation maps for Pascal-2012 dataset from kaggle. But the segmentation object folder contain just 2931 images, where can I access the other images? Original images are around 17k.

Thanks in advance

Hi author,

When I run the pre_train model as the instruction, why my program raised error:
(base) MacBook-Pro-cua-Do:Contour-Detection-Pytorch-master doxuanlong$ python3 ContourDetector.py a_58_2.jpg
Killed: 9

All the thing I changed is that I want to save the image instead of showing the image, so I changed:
plt.imshow(res)
plt.close()

to become
plt.imsave('test.jpg',res)

Thanks