Fix typo

README.md

@@ -23,7 +23,7 @@ To decode, the authors employ so-called decoder queries, which allow to flexibly
 
 <small> Perceiver IO architecture.</small>
 
-As the time and memory requirements of the self-attention mechanism don't depend on the size of the inputs, the Perceiver IO authors can train the model on raw pixel values, by concatening a pair of images along the channel dimension. 
+As the time and memory requirements of the self-attention mechanism don't depend on the size of the inputs, the Perceiver IO authors can train the model on raw pixel values, by concatenating a pair of images and extracting a 3x3 patch around each pixel. 
 
 The model obtains state-of-the-art results on important optical flow benchmarks, including [Sintel](http://sintel.is.tue.mpg.de/) and [KITTI](http://www.cvlibs.net/datasets/kitti/eval_scene_flow.php?benchmark=flow).