Performance on DARN
ChiWeiHsiao opened this issue · comments
Hi, I trained the ASENet_V2 on the DARN dataset with the default arguments in README (python main.py --model ASENet_V2 --decay_rate 0.9 --step_size 3 --epochs 50
), but it seems that the performance is inferior than that reported in Table 2.
Could you give me some suggestions about how to train the model (ex. hyperparameter settings)?
@ChiWeiHsiao Performance reported in Table 2 is obtained by training ASENet on DARN dataset for up to 200 epochs. In my later research, I modified the ASEN network to accelarate its training process on FashionAI dataset.
I got a 53.59% mAP on DARN dataset of ASENet_V2 with an additional linear layer at the end of the network. I suggest you add a linear layer and try again!
Looking forward to your further feedback!