lucasjinreal / Shadowless

A Fast and Open Source Autonomous Perception System.

Shadowless

The more fast your are, the more shadowless you got.

Shadowless is a new generation auto-drive perception system that feel things only in vision(more features maybe add in). We building shadowless on the purpose of establish a fully intelligent and fast drive system.

The 3 main part of shadowless are:

• Detection: this part implements many state-of-art detection algorithms to detect objects;
• Segmentation: currently we seg only on objects;
• Lane Detect: this will tell vehicle where should to ride and avoid touch the lane in 2 sides.

Preparation

You need to do some setups to run Shadowless, you should download mxnet_ssd model from mxnet official examples repo. and you should download one video from Youtube or anywhere place it into: videos/ directory. then:

sudo pip3 install -r requirements.txt

Please do 2 steps before you start running python3 main.py:

• Download the pretrained model from here, currently you should download Resnet-50 512x512, directly download url is here, after downloaded untar it into mxnet_ssd/model dir;
• You should install mxnet with the newest version.
• Mask-RCNN backend for detection will release very soon

Run

To run Shadowless after you get all pre-requirements, you can simply do:

python3 main.py

this will start a Shadowless master process to serve camera inputs and do the perception jobs.

Contribute

So much welcome the community contribute your code to Shadowless, we are now need those features:

• Detection with SSD

• Detection with FasterRCNN

• Detection with RFCN

• Lane Segment using OpenCV

• Lane Segment with DeepLearning methods

• Distance estimate with Objects

• Speed estimate with moving objects

• Accelerate whole networks to a real-time speed.

Copyright

this work inspired by Jin Fagang, you should not spread this soft-ware witout any guarantee, please using this under Apache License.