search

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

CIFAR-10, ResNet 110 score too low? (~5.9%) #177

Open FredericGodin opened 7 years ago

FredericGodin commented 7 years ago

I'm trying to reproduce the results for ResNet-110 on CIFAR 10.

I changed the lr schedule to the following to mimick the warm-up epoch: if epoch == 1 then decay = 1 else decay = epoch >= 122 and 2 or epoch >= 81 and 1 or 0 end

However, now my scores are too low... I used the following command: CUDA_VISIBLE_DEVICES=0,1 th main.lua -dataset cifar10 -nGPU 2 -batchSize 128 -depth 110 -netType resnet

I did 2 runs: run 1: best top1=6.04, final top1=6.04 run 2: best top1=5.77, final top1=5.89

The blog post reports top1 scores of ~6.7%

Any ideas?

Juicechen95 commented 6 years ago

maybe it is not the accuracy, it is the error.......accuracy = 100 - error

JaeDukSeo commented 6 years ago

Yeah it seems like an error rate