Low Accuracy on ilsvrc12_val using VGG16 pre-trained network.

[maxkohlbrenner]

Hi, according to https://keras.io/applications/#vgg16, the vgg16 pre-trained model has a validation top-1 accuracy of 71.3%. I deployed the network with the given weights and preprocessing function and arrived at a top-1 accuracy of 64.2% on the ilsvrc12 validation set. Can anyone confirm the reported accuracy of the pre-trained network?

Thank you very much for your help! Cheers, Max

[taehoonlee]

@maxkohlbrenner,

I will share my inference codes in the near future. Key recipes for ImageNet are the following:

• Down-sampling should not be performed as a square. Adjust shorten side as 256 is the right way. In other words, a resized image is either one of (256 X W) or (H X 256).
• The down-sampling scheme is pretty important. I evaluated all the models with cv2.resize(img, (h, w), interpolation=cv2.INTER_CUBIC), while the keras default is img.resize((w, h), pil_image.NEAREST). You can try pil_image.BICUBIC or opencv2 like me.
• After the down-sampling, you should crop the 224x224 center region from an (256 X W) or (H X 256) image.

[maxkohlbrenner]

Thank you very much @taehoonlee! This solved my problem.