Huge percentage error for a 200 imagenet classes - Githubissues

facebookarchive / fb.resnet.torch

Torch implementation of ResNet from http://arxiv.org/abs/1512.03385 and training scripts

Other

2.29k stars 664 forks source link

Huge percentage error for a 200 imagenet classes #175

Closed mh3081995 closed 7 years ago

mh3081995 commented 7 years ago

I'm fine tuning resnet-50 using the following command

th main.lua -resetClassifier true -nClasses 200 -retrain resnet-50.t7 -nEpochs 20 -data ~/Desktop/data/ -save ~/Desktop/ -batchSize 32 -LR 0.0001

and i have already changed: model:add(nn.Linear(nFeatures, 1000)) -> model:add(nn.Linear(nFeatures, 200)) in https://github.com/facebook/fb.resnet.torch/blob/master/models/resnet.lua#L128

and I'm getting top 1 96% in the first epoch and for epoch 15, i got top1 77%

is that normal?