search

GauravBh1010tt / MD-DETR

Continual Detection Transformers (ECCV'24)
4 stars 1 forks source link

Something about the result #1

Open Xuxiaoxiaohaha opened 5 days ago

Xuxiaoxiaohaha commented 5 days ago

Hi,good works!!!

The results of saved checkpoints you provided are slightly different from the paper.May I ask why this is?

I guess it's because you used different seeds for the experiment and took the mean in the paper. If that's the case, can you let me know which seeds you took so that I can better reproduce them? Or perhaps for some other reason?

Thanks!!!

GauravBh1010tt commented 3 days ago

Hi,

Thank you for reaching out. I just noticed that the checkpoints are from the model I trained during the rebuttal. The performance is similar to the original model, but I see some differences in T2, T3, and T4. I deleted most of the checkpoints as they took up a lot of space. I can try reproducing the results and share it with you but it may take a while.

Best, Gaurav