search

microsoft / arcticseals

A deep learning project in cooperation with the NOAA Marine Mammal Lab to detect & classify arctic seals in aerial imagery to understand how they’re adapting to a changing world.
MIT License
33 stars 14 forks source link

Adding statistsics and image loader. #10

Closed CIPop closed 6 years ago

CIPop commented 6 years ago

Using Pandas to explore the data. Adding sample code to load the color/ir images.

pbaer commented 6 years ago

I notice this is added to src/SealDetectionCNN - did you mean src/SealDetectionRCNN? Or is it meant to be independent from that sub-project?

CIPop commented 6 years ago

did you mean src/SealDetectionRCNN? Or is it meant to be independent from that sub-project?

I don't plan to use RCNN which would give us bounding boxes around seals from the larger image. Instead I'm following the approach in the issues:

  1. Use an existing CNN (VGG16 / inception, etc) to detect seals based on the processed data from NOAA (#5). My approach uses color data only.
  2. Create a better registration algorithm between IR and Color (#2)
  3. For new images:
    • Register IR / Color
    • Detect hotspots (ML or signal analysis)
    • Run the trained CNN
CIPop commented 6 years ago

@pbaer @Marcel-Simon before we push this one, let's all meet and discuss some folder structure. We could do one based on the issue structure:

  1. Image Registration
  2. Hotspot detection
  3. Hotspot classification
  4. Tooling (maybe specific to Azure, etc)