cant reproduce the paper performance

[bixiaopeng0]

Could you provide a training environment? thanks

[YinengXiong]

me too, only get about ~18 on Car AP40 using the config in the repo

[Chenchenwenyu]

Me too, is there any way to reproduce a better result?

[yexiguafuqihao]

The same to me, can reproduce the performance posted in the paper. The provided checkpoint can't be loaded to initialized the model for inference.

[ZrrSkywalker]

My apology for this. This repo contains only an intermediate version of MonoDETR. Our paper, MonoDETR, is still under review, but has been intentionally plagiarized by several times in character level, submitting to NeurIPS, CVPR, and other conferences. Given this, we plan to release the complete code after our paper been accepted. Thanks for your understanding.

[pummi823]

Hello, i see your article has been accepted by CVPR, will you release the complete code?~

[ZrrSkywalker]

@pummi823 Hi Thanks for your interest. Our paper is still under review :).

[ZrrSkywalker]

Thanks for your waiting @yexiguafuqihao @YinengXiong @bixiaopeng0 @pummi823 The more stable code of MonoDETR has been released.

[KotlinWang]

Me too, is there any way to reproduce a better result?

Hello, I am reproducing the training results of the paper, but when I use a single 3090 for training according to the requirements in the paper, it prompts that the CUDA memory is out. What is the reason? Looking forward to your reply, thank you!