fix some tools and tmp numpy version

README.md

@@ -40,6 +40,10 @@ Our runner directly saves the odometry estimate file in KITTI format (homogenous
 
 Our pipeline is `anytime realtime`! You can play with parameters `num_keyframes` and `num_cores` and, if you have enough _computation capacity_, we suggest increasing these (we run demo/experiments with `num_keyframes=16` and `num_cores=16`).
 
+## Data associtation and registration tools
+If you want to use our MAD-tree to perform nearest neighbor or use MAD-ICP to perform registration between two point clouds, <b>[here few easy examples](https://github.com/rvp-group/mad-icp/tree/main/mad_icp/apps/utils/tools/README.md)</b>.
+
+
 ____________________________________________________________________
 
 # Building from source

mad_icp/apps/utils/tools/README.md

@@ -1,23 +1,24 @@
-<div align="center">
-    <h1>Data association and registration tools</h1>
-    <br />   
-</div>
 
-# Usage
+# Data association and registration tools
+### Data association
 `nn_search.py` allows computing the data association between either:
 - a query point and a reference cloud;
 - a query cloud and a reference cloud.
 
 For each query point, the nearest-neighbor search returns the corresponding reference point with the estimated normal.
-Notice that the pybind of the tree build method has `b_max` parameter set to `1e-5` by default (check `src/pybind/tools/pymadtree.cpp`). A MAD tree built with such a parameter has exactly one leaf for each point of the original cloud, thus providing the most dense possible reference.
-Indeed, the total matching error printed by running `nn_search.py` as it is is zero. Highering `b_max` in the tree building leads to a shallower tree with fewer leaves, thus making the reference less dense than the query cloud, consequently increasing the total matching error.
+Notice that the pybind of the tree build method has `b_max` parameter set to `1e-5` by default (check `src/pybind/tools/pymadtree.cpp`). A MAD-tree built with such a parameter has exactly one leaf for each point of the original cloud, thus providing the most dense possible reference.
+Indeed, the total matching error printed by running `nn_search.py` as it is is zero (because query and reference cloud are the same). Highering `b_max` in the tree building leads to a shallower tree with fewer leaves, thus making the reference less dense than the query cloud, consequently increasing the total matching error.
 To run this module:
 ```bash
 python3 nn_search.py
 ```
-
+### Registration
 `mad_registration.py` registers a query cloud on a reference cloud, providing the estimated transformation between them. This module also allows us to visualize the cloud alignment at the end of each iteration, and the data reassociations between the two.
 You can run this module with the option `--viz` for toggling visualization.
 ```bash
 python3 mad_registration.py --viz
 ```
+
+<div align="center">
+        <a href=""><img src="mad-registration.gif?raw=true"/></a>   
+</div>

mad_icp/apps/utils/tools/mad_registration.py

@@ -79,7 +79,7 @@ def main(viz: Annotated[bool, typer.Option(help="if true visualizer on", show_de
     query_pcd.paint_uniform_color([1, 0, 0])  # red
 
     vis = o3d.visualization.Visualizer()
-    vis.create_window()
+    vis.create_window(window_name="MAD-registration", width=1280, height=720, visible=True)
     vis.add_geometry(ref_pcd)
     vis.add_geometry(query_pcd)
 

pyproject.toml

@@ -25,7 +25,7 @@ classifiers = [
 ]
 dependencies = [
     "PyYAML",
-    "numpy",
+    "numpy==1.26.4",
     "rosbags",
     "open3d",
     "matplotlib",