From ee91af23f5651f4a00c828cf8a353c7a24cc6c96 Mon Sep 17 00:00:00 2001 From: Guillermo Date: Fri, 15 Nov 2024 14:03:25 -0300 Subject: [PATCH] Clarifications --- README.md | 4 +++- 1 file changed, 3 insertions(+), 1 deletion(-) diff --git a/README.md b/README.md index 9316bae..969fada 100644 --- a/README.md +++ b/README.md @@ -48,5 +48,7 @@ println("Perceptual quality score: ", score) - `pretraining`:`String` - defines which pretraining setup will be used. By default set to "DISTS". It supports "Flux" an "Mixed". "Flux" loads the 16 layers of VGG without modification. "Mixed" loads the convolutional layers of VGG without modification, but the pooling layers weights use the weights from `net_params.mat` ## Comments -- (AFAIK) Julia does not have an implementation of the DISTS metric, this "package" should do the trick. However, Flux implementation seems to produce different results with respect to PyTorch and MATLAB despite having the same weights. +- (AFAIK) Julia does not have an implementation of the DISTS metric, this "package" should do the trick. +- Flux implementation seems to produce different results with respect to PyTorch and MATLAB despite having the same weights. The results are similar and consistent, though. +- The DISTS score between two equal images is near to zero. - Pooling layers from DISTS are implemented as DepthwiseConvolutions. \ No newline at end of file