Hello, I used your resnet50 with ibna added, and the performance on the mixed data set is significantly reduced. What is the reason for this? The data set is multiple person reID data sets

[CWF-999]

Hello, I used your resnet50 with ibna added, and the performance on the mixed data set is significantly reduced. What is the reason for this? The data set is multiple person reID data sets

[XingangPan]

@wfchenwf I am not able to diagnose this issue given limited information, but you may refer to this repo: https://github.com/michuanhaohao/reid-strong-baseline, where they obtain promising results in person reID.

[asdfqwer2015]

Hi, I met a similar issue, have you solved this issue? @wfchenwf I used a mobilenet backbone in detection task, and added IN(with affine) after first conv and added IN in 3rd conv_dw(conv-bn-relu-conv-in-relu) block refer to resnet50-ibnb. one of the mixed dataset is VisDrone2019, another one is our own drone dataset and the image quality is lower than VisDrone. Although I trained detection task from scratch since IBN arch, but it converged to a much higher trainloss and valloss than BN only model trained from scratch. What did I missed? @XingangPan Thanks.