Self-Ensembling Single Shot Detector (SESSD)

SE-SSD is a 3d-object-detection model: https://github.com/Vegeta2020/SE-SSD

detection-radius: 50[m], benchmark: TITAN GTX, RTX 3060Ti
Average inference time(180-degree): 30[ms] ≈ 30[fps]
Average inference time(360-degree): 65[ms] ≈ 15[fps]

Other Work

Usage

ignore the repo src just pull the docker image and follow the usage guide

$ docker pull loaywael/sessd:ros-torch1.9-cuda111

creating a container from this image.

$ docker run -itd --name SESSD_ROS --gpus all --net host --ipc host \
    -v <host-shared-path>:/shared_area loaywael/sessd:ros-torch1.9-cuda111
$ docker exec -it SESSD_ROS bash

run the demo node

# 1st option using roslaunch
# default kitti-bag for arl-bag set arl:=true
$ roslaunch se_ssd detect_3d_objects.launch kitti:=true weights_path:=<abs-path>

[evaluation]

Reproduced the model results

Evaluation official_AP_11: car AP(Average Precision)@0.70, 0.70, 0.70:
bbox AP:98.72, 90.10, 89.57
bev  AP:90.61, 88.76, 88.18
3d   AP:90.21, 86.25, 79.22
aos  AP:98.67, 89.86, 89.16
car AP(Average Precision)@0.70, 0.50, 0.50:
bbox AP:98.72, 90.10, 89.57
bev  AP:98.76, 90.19, 89.77
3d   AP:98.73, 90.16, 89.72
aos  AP:98.67, 89.86, 89.16

Evaluation official_AP_40: car AP(Average Precision)@0.70, 0.70, 0.70:
bbox AP:99.57, 95.58, 93.16
bev  AP:96.70, 92.15, 89.75
3d   AP:93.75, 86.18, 83.51
aos  AP:99.52, 95.28, 92.69
car AP(Average Precision)@0.70, 0.50, 0.50:
bbox AP:99.57, 95.58, 93.16
bev  AP:99.60, 95.92, 93.42
3d   AP:99.59, 95.86, 93.36
aos  AP:99.52, 95.28, 92.69

[to-do]

• model inference script.
• integrate with ROS
• support custom datasets
• support 360 degree detection
• support multi-class prediction