qdata / c-tran Goto Github PK

I am very interested in your work. Can you provide instructions for the operation of the CUB dataset?
Thank you very much!

Thanks for sharing the code!
Could you provide the label name of the VG500 subset and the download link of the dataset?

Hi,
Thank you for the code and the great work. I have two questions regarding your work. First how many number of epochs your model needs to train successfully? I am running your model on my dataset however it is not learning till 10 epochs therefore I want to know how many epochs your model needs. Secondly do you also have trained models as transformers need a lot of resources for training. Therefore if you can provide trained models that would help me a lot. Thank you for your time.

Thank you very much for your work, but I am confused about some code of your dataset, especially the coco type. Your code is not regular COCO dataset, and you did not provide relevant code to convert to your format, which made it very difficult for me to migrate to my own dataset.

Hello, we are unable to access the dataset download link you provided because you are using the school's online drive and we cannot access the edu website

system can not find GPU devices.... is there something wrong with the code？

@jacklanchantin @qiyanjun @jakegrigsby

How to implement C-Tran on custom dataset.

I have the information like
img_123.jpg - dog, cat, lion
img_124.jpg - cat,lion
etc...

Is there some specific format I need to follow to train with C-Tran

Dear author,

Thanks very much for your interesting work!
I wonder why the args "--grad_ac_step" is set as 2 for VOC2007?

From your code, I see that if you set "--grad_ac_step" as 2, a minibatch images (16 images) will not be used for optimization.
Can you help me solve this issue?

Thanks very much!

Hi,
I want to run the model by freezing the backbone I am getting the following error in main.py line 56:
'CTranModel' object has no attribute 'module'. Did you mean: 'modules'?

When I change module to modules then I get following error on same line 56 in main.py:
AttributeError: 'function' object has no attribute 'backbone'

Could you please help me regarding this?

Can you provide the content in "train.data" and "val_test.data" dirs? I have the coco images and annotations already according to the original coco website.

how to train C-Tran for Multi-class possibilities.

What I mean is
pug, husky, dolmination (outputs one label out of 3 possibilities)
cat/ not a cat (outputs one label out of 2 possibilities)

Excuse me,
I am very interested in your work. But the End2EndModel method is not found in this project. And MAP didn't live up to expectations.
I'm running on Windoes
Thank you very much

Thank you very much for your work, but I am confused about some code of your dataset, especially the coco type. Your code is not regular COCO dataset, and you did not provide relevant code to convert to your format, which made it very difficult for me to migrate to my own dataset.

Excuse me, could you please provide the mAP of VOC2007? I found it was not mentioned in the paper.

Thank you for your wonderful project!

I have been doing research with your C-Tran project, and conducting training with several autonomous driving dataset, such as nuScenes and METEOR.

It seems that when the number of images is larger than 100,000, all of the images tend to have almost same predictions. For example, [car: 0.96, pedestrian: 0.01, truck: 0.48] for all of the images, even though labels of different images are different. However, when I shrink the number of images to 3000 or less, different images will have different predictions of labels.

May I ask why this problem happens? Is it because I failed to apply your network correctly, or because the network with Transformers struggled to output satisfying predictions with large data?

Thank you for your reading, and I look forward to your reply.

I tried running your codes on JupyterLab, but there is a persistent error popping up every time I run it according to the steps you mentioned in README.md, and I'm unable to solve it. I have attached a screenshot of the same. I would be obliged if you could help me understand or solve the problem.

Thank you for your exciting work, I still have some detail implementation problems and would like to try some more ideas.
So hoping for your code! (●'◡'●)

Hi, I want to use C-Tran code. How to inference this code?
Could you explain how to inference this code after training?

Dear author,

Thanks for your time!

I have noticed that their is an argument "--pos_emb", so how to use it, If it will improve performance?
I found if I call it when training, it will raise an error.

Meanwhile, I wonder if this AAAI2021 paper can be compared with your paper? It seems these two papers are using the same setting.
https://ojs.aaai.org/index.php/AAAI/article/view/17098/16905

I tried running with VOC2007 dataset and worked but while I am trying to implement it with my dataset I am not able to run it due to missing annotations files, Is it mandatory to generate them or there is any way to ignore them and run.

My Dataset has only images and a CSV file with an image id and label column.

File "D:\Graduation project\归档\C-Tran-main\run_epoch.py", line 30, in run_epoch
    for batch in tqdm(data,mininterval=0.5,desc=desc,leave=True,ncols=50):
RuntimeError:
        An attempt has been made to start a new process before the
            if __name__ == '__main__':
                freeze_support()
                ...

        The "freeze_support()" line can be omitted if the program
        is not going to be frozen to produce an executable.

The above error occurs when I use this code to execute this program

Hi, I want to train on my custom dataset, but I cannot find any information on how one can do it. Can you provide a README or any form of documentation where I can look it up? Or can you help me understand and implement it myself?

output = self.output_linear(label_embeddings)
diag_mask = torch.eye(output.size(1)).unsqueeze(0).repeat(output.size(0),1,1).cuda()
output = (output*diag_mask).sum(-1)

I'm wondering what this diag_mask is playing in output value? Why should the output multiple a diagonal matrix? Your kind reply is very appreciated!

I am interested in your project!
I want to reuse the code, would you please add an open source license?