diff --git a/README.md b/README.md index 93d19851fe5e1b220afdec1b82f084a7a30ae592..648eea6328d603f1f32811705a933f65433afefe 100644 --- a/README.md +++ b/README.md @@ -56,6 +56,42 @@ color and dense LiDAR images for training the detec-tion network as shown in Fig  Fig. 1. The proposed (a) Color-based (b) Sparse LiDAR-based (c) Dense LiDAR-based and (d) Color and dense LiDAR based frameworks. + +#Qualitative Results +Fig.2 and Fig.3 illustrate four example detection results +from test KITTI dataset by the proposed frameworks with +Faster R-CNN and SSD detectors, respectively. The detection +results of the proposed fusion frameworks show that these +are able to detect targets more efficient than other proposed +frameworks. Note that because the early fusion framework can integrate information from both color and dense depth images. +The fusion frameworks successfully detected the size/location +of the bounding boxes. In the third and fourth examples, our +fusion framework has detected ”Pedestrians” and “Cyclist” +that other frameworks have missed. Moreover, the fusion +framework is able to detect small objects with a few pixels +as shown in Fig.2 (E) and many of them are detected by +our framework. It shows the generalisation capability of the +proposed framework and indicates its potentials in executing +2D object detection in real situations beyond a pre-designed +dataset. + + + + +Fig. 2. Qualitative results of the proposed frameworks with Faster R-CNN on four example images from test KITTI dataset. The first row of images is the +ground truths on input color images. The second is the color-based baseline framework. The third and forth rows of images are the detection result of two +uni-modals on sparse and dense depth images, respectively. The last row illustrates the detection result of multi-model framework on color and dense depth +image. + + + + + +Fig. 3. Qualitative results of the proposed frameworks with SSD on four example images from test KITTI dataset. The first row of images is the ground +truths on input color images. The second is the color-based baseline framework. The third and forth rows of images are the detection result of two uni-modals +on sparse and dense depth images, respectively. The last row illustrates the detection result of multi-model approach on color-dense depth image. + + # References 1. F. Farahnakian, and J. Heikkonen, “Fusing LiDAR and Color Imagery for Object Detection using Convolutional Neural Networks”, The 23th edition of the IEEE International conference on information fusion