search

heitorrapela / HalluciDet

[WACV2024] HalluciDet: Hallucinating RGB Modality for Person Detection Through Privileged Information (Accepted at WACV 2024 and LatinX@CVPR2024 Extended Abstract)
MIT License
19 stars 0 forks source link

Inquery the full validation results #1

Closed XueZ-phd closed 7 months ago

XueZ-phd commented 7 months ago

Dear author,

This is a good work! Therefore, I would like to cite this paper in my manuscript.

Could you provide more detailed results (i.e. AP50, AP75, AP50:95) on LLVIP and FLIR datasets? In this case, we can report them directly.

All the best!

heitorrapela commented 7 months ago

Hello @XueZ-phd, thanks for your interest in our work.

Unfortunately, for this work specifically (main paper and support material for WACV2024), we just reported the AP50, and here we run five seeds for each model. But for the entire study on the HalluciDet with three seeds for all APs (i.e., AP50, AP75, AP50:95), I would recommend our follow-up work (ModTr), were we re-run all baselines, and we improved the image-to-image baselines.

The HalluciDet in ModTr paper reported over three seeds and was in a new environment, but the results were a bit superior. The code for this new experiment was done over this repo. As this new work is still a pre-print, It can have some modifications for the camera-ready version, but for the HalluciDet results, you can trust. So, If you want to cite the HalluciDet, you can use our ModTr supp material results and add a footnote saying that these results were done over three seeds provided by the original code.

Full results for HalluciDet can be found on supp material: https://arxiv.org/pdf/2404.01492

image

Let me know if you have any additional questions,

Best Regards, Heitor Medeiros

XueZ-phd commented 7 months ago

Dear Heitor Medeiros,

Thanks for your prompt reply, it saved me a lot of time.

I will report the mean value of HalluciDet in my manuscript: AP50=90.07, AP75=51.23, and AP=57.78.

HalluciDet is really a great work. It has the interesting motivations and good results. I hope more researchers will cite it.

Best Regards, Xue Zhang

heitorrapela commented 7 months ago

Dear Heitor Medeiros,

Thanks for your prompt reply, it saved me a lot of time.

I will report the mean value of HalluciDet in my manuscript: AP50=90.07, AP75=51.23, and AP=57.78.

HalluciDet is really a great work. It has the interesting motivations and good results. I hope more researchers will cite it.

Best Regards, Xue Zhang

That sounds good. Thanks for your interest in our work.

I hope more researchers also like it,

Best Regards, Heitor Medeiros

heitorrapela commented 3 months ago

Hi, @XueZ-phd how is going with your manuscript? I hope it is going well. Just to update, for subsequent works if you are still interested in this line of research, our paper ModTr was accepted at ECCV2024, it might be interesting to look on it as well.

Best Regards, Heitor Medeiros

XueZ-phd commented 3 months ago

Hi,@heitorrapela. Congratulations!

Our manuscript has recently been accepted by the IEEE T-NNLS journal. Here.

I am very interested in your work. We will focus on it in our follow-up work.

Best Regards,

Xue Zhang

heitorrapela commented 3 months ago

Hi @XueZ-phd.

Congrats on your paper!

Thanks for the kind words,

Best Regards, Heitor Medeiros