diff --git a/README.md b/README.md
index cdc0be2..48e60ec 100644
--- a/README.md
+++ b/README.md
@@ -2,23 +2,7 @@
 
 The subject is available here: [Test Algo](./test_algo.pdf)
 
-## TODO
-- [ ] document
-- [X] dumb packing 
-- [ ] skyline
-- [X] add raygui
-- [X] resize box
-- [X] delete box in gui by clicking
-- [ ] wrap stb_rect_pack?
-- [X] change bbox api to origin + size
-- [X] dummy A2
-- [X] package
-
-##  Installation
-
-## Notes
-
-### Question 1
+## Question 1
 
 > As a preprocessing step of a second algorithm A2, we would like to combine all
 > the regions corresponding to the bounding boxes into a new image FRegions of
@@ -32,10 +16,13 @@ I looked up some solutions online and found a great article by David Colson:
 "[Exploring rectangle packing algorithms][2]". It gives a lot a references and
 compares different algorithms.
 
-I decided at first to implement his naive "row packer" to quickly have an
-implementation and try it out.
+I decided to implement his naive "row packer" to quickly have an implementation
+and try it out.
 
-### Question 2
+If I were to implement a more robust algorithm, I would choose the [Skyline][3]
+approach. It looks likes a good trade-off of quality and performance.
+
+## Question 2
 
 > A2 then takes as input F Regions and outputs new bounding boxes B′. We would
 > like now to compute the location of each of these new bounding boxes in F
@@ -59,7 +46,7 @@ recompute them by looking at pixel similarity between F and FRegions but it
 would be costly, and a bit wasteful since we already computed the bounding
 boxes.
 
-### Question 3
+## Question 3
 
 > We would like now to be able to provide to the algorithm A2 either the
 > region-based image or the initial image without transformation, which
@@ -72,10 +59,12 @@ support for a binary mask. In the same way A2 is suppose to ignore zeros valud
 in the region-based frame, it could be modified to ignore areas in an image
 where the mask is set to zero.
 
-## GUI
+## Installation
+
+### GUI
 
 In order to easily debug and better visualize the problem, I chose to implement
-a minimal GUI using [raylib][3].
+a minimal GUI using [raylib][4].
 
 You can build it with `./build-gui.sh` (you need to installed [raylib required
 libraries](https://github.com/raysan5/raylib/wiki/Working-on-GNU-Linux)).
@@ -83,7 +72,7 @@ libraries](https://github.com/raysan5/raylib/wiki/Working-on-GNU-Linux)).
 You can add, move, and resize boxes. Processing steps are triggered with
 buttons.
 
-## CLI
+### CLI
 
 A commandline sample is also available, in case the raylib library cannot be
 built, or if we need to benchmark performance.
@@ -113,4 +102,5 @@ To avoid name collision, I created my own namespace `freling`.
 
 [1]: https://en.wikipedia.org/wiki/Packing_problems#Packing_of_rectangles
 [2]: https://www.david-colson.com/2020/03/10/exploring-rect-packing.html
-[3]: https://github.com/raysan5/raylib
+[3]: https://www.researchgate.net/publication/221049934_A_Skyline-Based_Heuristic_for_the_2D_Rectangular_Strip_Packing_Problem
+[4]: https://github.com/raysan5/raylib
diff --git a/src/pack.cpp b/src/pack.cpp
index 9b922ea..c328919 100644
--- a/src/pack.cpp
+++ b/src/pack.cpp
@@ -28,6 +28,9 @@ void blit(const Frame& in_frame,
     }
 }
 
+// This packer is based on the "row packing" approach as described by David
+// Colson in his article:
+// https://www.david-colson.com/2020/03/10/exploring-rect-packing.html
 std::optional<Frame> pack(const Frame& in_frame,
                           const std::vector<BoundingBox>& bboxes,
                           std::vector<BoundingBox>& packed_bboxes) {