Add LaneNet evaluation

doc/06_perception/experiments/lanenet_evaluation/README.md

@@ -0,0 +1,38 @@
+# LaneNet Evaluation
+
+To see whether [LaneNet](https://github.com/MaybeShewill-CV/lanenet-lane-detection) from [Pylot](https://github.com/erdos-project/pylot) can be used for our purposes in Carla, a small evaluation s- imilar to the object detection experiment - was conducted.
+
+## Installation
+
+1. Create a `venv` with Python 3.7 (not newer!)
+2. Run `pip install -r requirements.txt`
+3. Set your `PYTHONPATH` like this:
+
+```shell
+export PYTHONPATH="${PYTHONPATH}:/PATH/TO/PROJECT/lanenet-lane-detection/lanenet_model"
+```
+
+## Inference
+
+To use the model on a certain image for demo purposes, the following command can be used:
+
+```shell
+python tools/test_lanenet.py --weights_path /PATH/TO/YOUR/CKPT_FILE_PATH 
+--image_path ./data/tusimple_test_image/0.jpg
+```
+
+This generates a set of plots, two of which are the original image with the recognized lanes drawn on top, as well as a mask of detected lanes (which are not similar).
+
+## Results
+
+As visible below, the masks show a better and less distorted result than the drawn image.
+For usage in our agent, the mask would probably be used anyway, so that is not a serious concern.
+
+The middle line is ignored everywhere and on the third and partly also on the fourth sample picture, the recognition of the lanes failed.
+
+Image             |  Mask
+:-------------------------:|:-------------------------:
+![1600-Image](assets/1600_lanes.jpg)  |  ![1600-Mask](assets/1600_lanes_mask.jpg)
+![1619-Image](assets/1619_lanes.jpg)  |  ![1619-Mask](assets/1619_lanes_mask.jpg)
+![1660-Image](assets/1660_lanes.jpg)  |  ![1660-Mask](assets/1660_lanes_mask.jpg)
+![1663-Image](assets/1663_lanes.jpg)  |  ![1663-Mask](assets/1663_lanes_mask.jpg)