update kitti eval

readme/GETTING_STARTED.md

@@ -50,12 +50,10 @@ And you can change to other network architectures and resolutions by specifying
 
 ### KITTI
 
-To evaluate the kitti dataset, first clone a third-party evaluation tool (if you haven't added `--recursive` when cloning this repo) and compile:
+To evaluate the kitti dataset, first compile the evaluation tool (from [here](https://github.com/prclibo/kitti_eval)):
 
 ~~~
-cd CenterNet_ROOT/src/tools/
-git clone https://github.com/prclibo/kitti_eval
-cd kitti_eval
+cd CenterNet_ROOT/src/tools/kitti_eval
 g++ -o evaluate_object_3d_offline evaluate_object_3d_offline.cpp -O3
 ~~~
 

readme/INSTALL.md

@@ -47,7 +47,7 @@ After install Anaconda:
 
     ~~~
     CenterNet_ROOT=/path/to/clone/CenterNet
-    git clone --recursive https://github.com/xingyizhou/CenterNet $CenterNet_ROOT
+    git clone https://github.com/xingyizhou/CenterNet $CenterNet_ROOT
     ~~~
 
 

src/tools/kitti_eval/README.md

@@ -0,0 +1,22 @@
+# kitti_eval
+
+`evaluate_object_3d_offline.cpp`evaluates your KITTI detection locally on your own computer using your validation data selected from KITTI training dataset, with the following metrics:
+
+- overlap on image (AP)
+- oriented overlap on image (AOS)
+- overlap on ground-plane (AP)
+- overlap in 3D (AP)
+
+Compile `evaluate_object_3d_offline.cpp` with dependency of Boost and Linux `dirent.h` (You should already have it under most Linux).
+
+Run the evalutaion by:
+
+    ./evaluate_object_3d_offline groundtruth_dir result_dir
+
+Note that you don't have to detect over all KITTI training data. The evaluator only evaluates samples whose result files exist.
+
+
+### Updates
+
+- June, 2017:
+  * Fixed the bug of detection box filtering based on min height according to KITTI's note on 25.04.2017.