diff --git a/docs/usage/normalization_methods.md b/docs/usage/normalization_methods.md
index a0d945f5..81f25a2a 100644
--- a/docs/usage/normalization_methods.md
+++ b/docs/usage/normalization_methods.md
@@ -17,8 +17,8 @@ Additionally, the normalized data can be scaled using scanpy's [sc.pp.scale](htt
    2. clr margin= 1, normalize within each feature's distribution, across all cells (column-wise, recommended for proteomics data)
 
     *If you come from R, please note that the [margins are transposed](https://images.hindustantimes.com/rf/image_size_630x354/HT/p2/2017/09/21/Pictures/_78c6a162-9e94-11e7-9c3b-8e901839ece0.JPG) in the Python and anndata world.*
+    <img src="https://github.com/DendrouLab/panpipes/blob/main/docs/img/clr_margins.png?raw=true" alt="img1" width="40%">
     
-    <img src="../img/clr_margins.png" width="40%">
 
 2. dsb using [muon's prot processing](https://muon.readthedocs.io/en/latest/api/generated/muon.prot.pp.html). This method is only applicable when you have raw 10x inputs (see [supported input files](https://panpipes-pipelines.readthedocs.io/en/latest/usage/setup_for_qc_mm.html#supported-input-filetypes)).