search

forresti / SqueezeNet

SqueezeNet: AlexNet-level accuracy with 50x fewer parameters
BSD 2-Clause "Simplified" License
2.17k stars 723 forks source link

Image Preprocessing for stated top5 accuracy #23

Closed fervorarc closed 6 years ago

fervorarc commented 7 years ago

What image preprocessing was performed on the imagenet images to achieve the stated feedforward top5 accuracy? E.g. resize uniformly to 256 at the smallest dimension, then center crop I've had a tough time figuring this out, and any help would be much appreciated. Many thanks!

Cospel commented 7 years ago

Hi, that will be great.

I cannot find this in the paper. Can authors please respond. I am trying to implement and train it from scratch in Tensorflow, but I am not sure what strategy did you used.

itdxer commented 6 years ago

FYI @fervorarc, @Cospel I found this information specified in caffe training file here: https://github.com/DeepScale/SqueezeNet/blob/master/SqueezeNet_v1.0/train_val.prototxt

  transform_param {
    crop_size: 227
    mean_value: 104
    mean_value: 117
    mean_value: 123
}