Updates

WW 29 (July) |

• Understand nuScenes dataset
• Short meeting with Santosh Meeting Notes

WW 30 (July) |

• Understand nuScenes dataset
• Find out the top fusion methods based on nuScenes

WW 31 (August) |

• List down methods with tightly coupled vs. without tightly-coupled

WW 32 (August) |

• Read one survey paper to get a better understanding of Radar data processing
• Implement one fusion method with the nuScenes dataset (Pt.1/2)
• Facing a lot of issues with dataset and project building

WW 33 (August) |

• Shortlisted fusion methods and datasets based on applicability on adverse weather
• Working project building issues

WW 34 (August) |

• Implement SAF-FCOS fusion method with the nuScenes dataset (Pt.1/2)
• Working on issues in the data preparation part
• Working project building issues

WW 35 (August) |

• Implement SAF-FCOS fusion method with the nuScenes dataset (Pt.1/2)
• Resolving issues in the data preparation part
• Put on training on the cluster (Pt.1/3)
• Implement one fusion method with the nuScenes dataset (Pt.2/2)

WW 36 (September) |

• Exams

WW 37 (September) |

• Break

WW 38 (September) |

• Back to Basics | PyTorch basics and concepts

WW 39 (September) |

• Back to Basics | Implemented Vision Transformer (classification) from scratch - LINK
• Back to Basics | Implemented YOLO v1 from scratch - LINK

WW 40 (October) |

• Back to Basics | Implemented custom 2D object detector LINK
• Back to Basics | Implement Point Cloud-based object classification (PointNet) LINK
• Learned FCOS
• Learned Focal loss, types of IoU loss (GIoU, DIoU, CIoU)
• Understand the architecture of SAF-FCOS
• Test results on the SAF-FCOS model

WW 41 (October) |

• Analysis of the SAF-FOCS model
• Read the HRFuser paper and understand the architecture well
• (if time permits) Read MT-DETR paper
• Replicate HRFuser model - Code Link
• Train HRFuser on the DENSE dataset
• Train HRFuser on the nuScenes dataset
• (if time permits) train a very basic baseline method with CR(L) based object detection - e.g. FasterRCNN

WW 42 (October) |

• Meeting with Prof. Houben
• Replicate HRFuser model - Code Link
• Train HRFuser on the DENSE dataset
• Train HRFuser on the nuScenes dataset
• Is it possible to fuse Lidar information to the SAF-FCOS model?
• Train Camera+Radar SAF-FCOS on DENSE dataset
• Compare SAF-FCOS and HRFuser results and architecture
• Detail analysis of results

WW 43 (October) |

• Submit a very first report draft for review
• Compare tightly coupled vs. simple fusion models
• Check how to convert a model from 3D object detection to 2D object detection (Update: HRFuser is doing 3D->2D)
  • Not required
• (if time permits) Train MT-DETR model
• (if time permits) train a very basic baseline method with CR(L) based object detection - e.g. FasterRCNN
• Train Camera+Radar SAF-FCOS on DENSE dataset
• Compare SAF-FCOS and HRFuser results and architecture
  • Not directly comparable because both are using different numbers of classes (Have to ask Santosh)

WW 44 (November) |

• Submit a very first report draft for review
• Compare tightly coupled vs. simple fusion models
• RUN MORE EXPERIMENTS
• Is it possible to fuse Lidar information to the SAF-FCOS model?
• (if time permits) train a very basic baseline method with CR(L) based object detection - e.g. FasterRCNN
• Train Camera+Radar SAF-FCOS on DENSE dataset
• Compare SAF-FCOS and HRFuser results and architecture
  • Not directly comparable because both are using different numbers of classes (Have to ask Santosh)

WW 45 (November) |

• Report writing

WW 46 (November) |

• Report writing | Draft 1

WW 47 (November) |

• Report writing and review | Draft 2

WW 48 (November) |

• Report writing and review