result

[lengmianlaotang]

Thank you very much for your work. I am unable to fully reproduce the results described in the paper using the provided checkpoint. The results for the car are very close to those reported in the paper, however, there are significant discrepancies for the cyclist, van, and pedestrian categories.I don't know where the problem lies.If you could reply, I would be extremely grateful.

[haooozi]

Hello, thanks for your this news. The results are reproduced on RTX 3070. We find that different devices may exhibit different results using the same environment. We attribute this phenomenon to the code. Please kindly refer to our new code of P2P, where the code is not sensitive to the device. 聂佳浩 ***@***.*** 签名由 网易灵犀办公 定制

…

---- Replied Message ---- From lengmianlaotang ***@***.***> Date 09/10/2024 16:47 To haooozi/GLT-T ***@***.***> Cc Subscribed ***@***.***> Subject [haooozi/GLT-T] result (Issue #3) Thank you very much for your work. I am unable to fully reproduce the results described in the paper using the provided checkpoint. The results for the car are very close to those reported in the paper, however, there are significant discrepancies for the cyclist, van, and pedestrian categories.If you could reply, I would be extremely grateful. image.png (view on web) image.png (view on web) image.png (view on web) — Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>