From 8d707cba97ca7c653cec43eb3e7d8fb8853e2776 Mon Sep 17 00:00:00 2001 From: Mike Walmsley Date: Mon, 25 Apr 2022 14:09:46 +0100 Subject: [PATCH] formatting error --- docs/data_notes.rst | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-) diff --git a/docs/data_notes.rst b/docs/data_notes.rst index 803ad11e..071f2447 100755 --- a/docs/data_notes.rst +++ b/docs/data_notes.rst @@ -34,7 +34,7 @@ See :ref:`the DECaLS guide ` for pedagogical details on h TensorFlow ........... -We provide pretrained weights for an EfficientNet B0 CNN (defined here) trained on the GZ DECaLS GZD-5 campaign volunteer votes. You can download them from from Dropbox `_. +We provide pretrained weights for an EfficientNet B0 CNN (defined here) trained on the GZ DECaLS GZD-5 campaign volunteer votes. You can download them `from Dropbox `_. - ``effnet_train_only_dr5_greyscale_tf`` is trained on the colour (3-channel grz) images shown to volunteers, but the images are averaged across bands before being input. - ``effnet_train_only_dr5_color_tf`` is identical but without averaging across bands. This approach was not used for the GZ DECaLS catalog to avoid bias, but may be useful for e.g. anomaly-finding. @@ -51,7 +51,7 @@ We provide pretrained weights for an EfficientNet B0 CNN (defined here) trained PyTorch ........ -We provide pretrained weights for EfficientNetB0, as with TensorFlow above, and for ResNet50. You can download them from from Dropbox `_. +We provide pretrained weights for EfficientNetB0, as with TensorFlow above, and for ResNet50. You can download them `from Dropbox `_. - ``efficientnet_dr5_pytorch`` and ``efficientnet_dr5_pytorch_color`` are trained similarly to the TensorFlow equivalents. - ``resnet_detectron_dr5_pytorch`` and ``resnet_detectron_dr5_pytorch_color`` are trained using the detectron2 implementation of ResNet50 (with unfrozen batch norm layers). See here.