The code with DETR

[ssnlyx]

Hello, sorry to bother you, I have modified detr according to your BatchFormerv2 code on deformable detr, the parameters are also according to your settings, but the effect does not seem to be very good, could you please release your code on detr?

[zhihou7]

Hi @ssnlyx,

I have released the code on detr. Before that, I shared with you earlier. The core part is here: https://github.com/zhihou7/detr/blob/master/models/transformer.py#L121. You should notice the difference between deformable-detr and detr, that is the batch dimension is different. Meanwhile, you should share the prediction modules (any modules after the batchformer in the pipeline, simese stream pipeline)

Feel free to post if you have further questions.

[ssnlyx]

Thank you so much for your prompt reply! May your research get better and better.

[zhihou7]

Thank you very much

[ssnlyx]

I'm really sorry to bother you, due to the reason of my device, batchsize can only be opened to 2, and I have to queue up to use the device. In fact, I would like to use your great research as part of my course design. Could you please release the weights after training on detr using batchformerV2?

[zhihou7]

Hi @ssnlyx,
Currently, I can not login to our HPC, where I store all my pre-trained models, due to some reasons. I will release the weights once I can access the HPC again.

Regards,

[zhihou7]

Hi @ssnlyx, I notice there is a model in the university cloud. However, I am not sure if this model is the DETR model. Here is the public link https://cloudstor.aarnet.edu.au/plus/s/9hKw4GHPBozbI2y.

Sorry for that I can not use our HPC until now. I can not use the HPC to verify the files.

Regards,

[ssnlyx]

Thank you so much！ I will try it ！