Not able to reproduce the same results on Resnet as reported in the paper

[MuscleGood]

Hi, the acc of trained ann resnet is 0.80 , but when I try to convert it into snn using this code , the converted snn acc is only 0.77.
In the fuse_norm_replace function of the snn which reaches 0.77 on acc , I only norm the bn1 and bn2 in a resnet block, while I didn't norm the bn layer which is used to down_sample the identity. For the bn1, I used the max_act of the current bn layer and the add layer of the last resnet block (the add layer is added specially to obtain the max_act).
Then I tried to norm the down_sample bn also (different tries, include using the max_act of current bn and the add layer in last block ), the acc has dropped a lot.
By the way, the result on VGG16 is reproduced acceptable, thank you for providing the code! expecting your reply, thank you in advance!

[Met4physics]

hi, which kind of resnet did you use? resnet34 or resnet 18? and at what timesteps your snn reaches 0.77?