Week 02

March 06

• ranges of frames of SD1 that camera is underwater

• SD1_01272 video: SD1_01272_013470.png to SD1_01272_014310.png

• SD1_01273 video: SD1_01273_008850.png to SD1_01273_010140.png, SD1_01273_029700.png to SD1_01273_032010.png

• CCTV competitors:

  • Pipe sleuth
  • Burgess and Niple

• SD1 presentation

  • send Deven the frames that are blurry
  • DNV validation and test score for each split 90-10 and 95-5 run using resnet50 (A50) and resnet101 (B101)
  • get the result on DNV model on SD1 images (average F2 score calculated on matching classes)

March 07

• Convert CCTV pipelines notebook to py to run on compute canada
• There were problems when running create_image_label using access database with images from group 4 and group 6 from 406 DNV PACP videos.
• SD1 CCTV
  • SewerVideoUploads folder
• Get the statistic on how many images have the continuous defect out of the label images for SD1, DNV, and COV.

March 09

• List of videos need to fix access database

  • SANMN02175_132.MP4 (have D as defect code)

• Create image labels v2 notebook

  • Might not include clean_row_with_dist_incr_grt_num() for SD1 data
  • need to add code to discard the frames when camera is lowering

• Review SD1 label extraction results from Pavan

  • The word manhole can appear at three different times:
    • at the beginning when inspector enters general information about the CCTV inspection
    • at the distance 0.0 ft
    • at the end of the survey
  • Deposits Settled Gravel was extracted as Deposits Settled Compacted

• SD1 used XP and XB to describe Collapse Pipe Sewer and Collapse Brick Sewer. The current PACP only has X for collapse. No descriptors and no modifiers used for collapse.

March 10

• list of corrupted blurred images from 406 DNV videos
  • SANMN00778_415_015330.png (have been fixed)