Reproducing the reported performance #3
Description
Hi, thank you for sharing the code for great work.
I have tried to reproduce the results for mini-imagenet using resent18 (I believe this is the only architecture supported) as follows,
python train.py --n_way 5 --k_spt 5 --net_arch resnet18 --net_domain mini --strategy map --feature_or_logits 1
python test.py -l_n mini_resnet18_sgd_map_0.0001_5N5K_logits
I have used both map and fb for strategy and also both 0 and 1 for feature_or_logits. Unfortunately, I've only got around 71% although the reported number is 80.98%. The gap is too big to think this is due to randomness. Did I do something wrong here? It would be highly appreciated if you can help me out to reproduce the results. Thank you.