marcelomata / ossdc-visionbasedacc

Discuss requirments and develop code for #1-mvp-vbacc MVP (see also this channel on ossdc.org Slack)

OSSDC-VisionBasedACC

Discuss requirments and develop code for #1-mvp-vbacc MVP (see also this channel on ossdc.org Slack)

We have a few initial demos videos here, showing how detection/tracking/segmentation/depth estimation algoritms can be used:

• An initial demo with Mask R-CNN (for object detection and instance segmentation) in Google Colaboratory with GPU acceleration (see more demos bellow):

  • https://github.com/OSSDC/OSSDC-VisionBasedACC/blob/master/image-segmentation/ossdc_matterport_Mask_RCNN_colaboratory.ipynb

  Open directly in Google Colaboratory: https://colab.research.google.com/github/OSSDC/OSSDC-VisionBasedACC/blob/master/image-segmentation/ossdc_matterport_Mask_RCNN_colaboratory.ipynb

• An initial demo with SfMLearner (for depth and ego-motion estimation from monocular videos) in Google Colaboratory with GPU acceleration:

  • https://github.com/OSSDC/OSSDC-VisionBasedACC/blob/master/object_detection/SfMLearner_demo.ipynb

  Open directly in Google Colaboratory: https://colab.research.google.com/github/OSSDC/OSSDC-VisionBasedACC/blob/master/object_detection/SfMLearner_demo.ipynb

• Super cool! Now we can run this for free in the cloud with GPU acceleration at about 15FPS on 720p Youtube video, try this IPython example in Google Colaboratory, updated with live streaming of processed output to ffplay:

  • https://github.com/OSSDC/OSSDC-VisionBasedACC/blob/master/object_detection/ossdc_vbacc_object_detection_notebook_colaboratory.ipynb

  Open directly in Google Colaboratory: https://colab.research.google.com/github/OSSDC/OSSDC-VisionBasedACC/blob/master/object_detection/ossdc_vbacc_object_detection_notebook_colaboratory.ipynb

• Update demo code is here (see run demo scripts):

  • https://github.com/OSSDC/OSSDC-VisionBasedACC/tree/master/object_detection

• Learning to Fly by Driving. Training a drone to fly from monocular camera images and associated steering commands from a car. See the paper.

  • https://github.com/OSSDC/OSSDC-VisionBasedACC/tree/master/steering

.

• SSD Tensorflow based car detection and tracking demo for OSSDC.org VisionBasedACC PS3/PS4 simulator

  • The code for the demo is here (see ossdc scripts):
    • https://github.com/OSSDC/SSD-Tensorflow/tree/master/notebooks
  • It can ran in real time on local videos or network streams, including directly on YouTube videos (using youtube-dl program), like I described here:
    • https://github.com/OSSDC/SSD-Tensorflow/commit/26fa7eea5155ac3989476936628e62be3d773b95

• Real time YOLO detection in OSSDC Simulator running TheCrew on PS4 30fps

  • The code for the demo is here:
    • https://github.com/OSSDC/darknet
  • See here instructions on how to run it:
    • https://github.com/OSSDC/darknet/commit/dc76e5690c753ce093417e096313b390b5d5c2ae

It can be noted in the videos above that SSD Tensorflow is more accurate than the Yolo version.

Here are a few articles with more details about VBACC and OSSDC:

• How OSSDC was born:
  • SDC From 0 to 60 (*) in 4 weeks
  • Why do we need Open Source Self Driving Car development and how to get started
• A few updates on AI and Self Driving Cars
• Understand instead of Memorize
• What is next in OSSDC.org?
• What about putting a computer vision processor on the camera or sensor platform itself?
• Learn by example
• Get ready to Race AI with us at OSSDC.org
• Live Visual Speed Recognition at 5000FPS - OSSDC PS3/PS4 Simulator running GT5 on PS3

We use docker to easily run the OSSDC code, check these docker instructions:

• https://github.com/OSSDC/self-driving-car-1

For a consolidated view about OSSDC platform and related work, check the hacking book we are creating here:

• OSSDC Hacking Book