lass

Landmark Segmentation Suitcase

 ██▓    ▄▄▄        ██████   ██████
▓██▒   ▒████▄    ▒██    ▒ ▒██    ▒
▒██░   ▒██  ▀█▄  ░ ▓██▄   ░ ▓██▄
▒██░   ░██▄▄▄▄██   ▒   ██▒  ▒   ██▒
░██████▒▓█   ▓██▒▒██████▒▒▒██████▒▒
░ ▒░▓  ░▒▒   ▓▒█░▒ ▒▓▒ ▒ ░▒ ▒▓▒ ▒ ░
░ ░ ▒  ░ ▒   ▒▒ ░░ ░▒  ░ ░░ ░▒  ░ ░
  ░ ░    ░   ▒   ░  ░  ░  ░  ░  ░
    ░  ░     ░  ░      ░        ░

Requirements

ToDo

Network

• K-NN loss (zhou)
• direction loss (zhou)
• Pnp

Data Processing

• MDS for embedding generation (zhou)
• over-segmentation algorithm (huang)
• image segmentation (zhou)
• dynamic objects removal (huang)
• foreground/background objects annotation (huang)
• voting center generation (huang)
• code integration from previous(huang)

Experiments

• landmark segmentation on 2D3DS (zhou)
• download 7Scenes (zhou)
• run RobotCar test on visuallocalization benchmark (zhou)
• select a scene in 2D3DS for demo (zhou)