diff --git a/.gitlab-ci.yml b/.gitlab-ci.yml
index 4c7c2deae3ee6a73758a2b7634fb121dcf4075e1..2cc931445263f8ba33e2037c597a95e5470010be 100644
--- a/.gitlab-ci.yml
+++ b/.gitlab-ci.yml
@@ -19,8 +19,7 @@ variables:
   DOCKERHUB_BASE: mdl4eo/otbtf
   DOCKERHUB_IMAGE_BASE: ${DOCKERHUB_BASE}:${OTBTF_VERSION}
   CPU_BASE_IMG: ubuntu:22.04
-  GPU_BASE_IMG: nvidia/cuda:12.1.0-devel-ubuntu22.04
-
+  GPU_BASE_IMG: nvidia/cuda:12.0.1-cudnn8-devel-ubuntu22.04
 workflow:
   rules:
     - if: $CI_MERGE_REQUEST_ID || $CI_COMMIT_REF_NAME =~ /master/ # Execute jobs in merge request context, or commit in master branch
diff --git a/README.md b/README.md
index 5529823f5c6af231284453562411469e05d4abb1..f4eac4f2e1a8d3fd1af3edaa7a62066f513e05ef 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,7 @@
 # OTBTF: Orfeo ToolBox meets TensorFlow
 
 <p align="center">
-<img src="images/logo.png" width="160px">
+<img src="doc/images/logo.png" width="160px">
 <br>
 <a href="https://gitlab.irstea.fr/remi.cresson/otbtf/-/releases">
 <img src="https://gitlab.irstea.fr/remi.cresson/otbtf/-/badges/release.svg">
diff --git a/RELEASE_NOTES.txt b/RELEASE_NOTES.txt
index e0ac4a4eaebbea499e0464edacc49489c1e3ebdd..fd30467cc836e5b3d75cbedbc891ad9527215697 100644
--- a/RELEASE_NOTES.txt
+++ b/RELEASE_NOTES.txt
@@ -1,4 +1,4 @@
-Version 4.0.0alpha (4 apr 2023)
+Version 4.0.0 (5 apr 2023)
 ----------------------------------------------------------------
 * Big improvement of the documentation:
   - Re-structure the entire doc, remove deprecated stuff, etc.
@@ -7,8 +7,7 @@ Version 4.0.0alpha (4 apr 2023)
   - Add a new section on the python API (end-to-end tutorial)
   - A lot of new sections: distributed training, etc...
 * Refactoring all the python classes to enforce pep8
-* Ubuntu version: 22.04
-* Cuda version: 12.1.0
+* Base images: ubuntu:22.04 (cpu), nvidia/cuda:12.0.1-cudnn8-devel-ubuntu22.04 (gpu)
 * Tensorflow version: 2.12.0
 * Fixed Tensorflow error "Cannot register 2 metrics with the same name" + new test
 * Faster CI build thanks to bazel remote cache
diff --git a/doc/app_training.md b/doc/app_training.md
index 5ea6efaab7452e71744cfa845e15ed969f983086..eb3f4708f37121d5f63181eb2c2b08a783da9df9 100644
--- a/doc/app_training.md
+++ b/doc/app_training.md
@@ -1,11 +1,13 @@
-!!! Warning
+!!! Info
 
 	This section is for educational purposes. No coding skills are required,
 	and it's easy to train an existing model built with the Tensorflow API
-	v1. To have a full control over the model implementation and training 
-	process, the Tensorflow API v2 with Keras is the way to go.
-    If you are interested in more similar examples, please read the 
+	v1. If you are interested in more similar examples, please read the 
     [Tensorflow v1 models examples](reference/otbtf/examples/tensorflow_v1x/__init__.md)
+    However, to have a full control over the model implementation and training 
+	process, the [Tensorflow API v2 with Keras](api_tutorial.html) is the 
+    recommended approach (see also 
+    [distributed training](api_distributed.html)). 
 	
 ## TensorflowModelTrain