hbwater_cameratrap_pheno

building a pipeline for processing HB Water aquatic camera trap photos for machine learning, with hopes to support further efforts to produce models from similar datasets.

Project Steps

1. image data re-organization
2. ROI
3. project file organization, structure, and primary host - GDrive? Server?
4. created inverted masked images
5. classify pixels in image using image annotator
6. run random forest model
7. create data product: markdown
8. create data product: application

Current Objectives

Tasks	People	Step
continue debugging the GDrive image copy/rename/organization scripts	BG, HS	1
figure out how to show mask on consecutive images (clicking thru all the images from a particular watershed year)	HO, BG	2
allow user to, while clicking through images, stop and create a new mask associated with the date of that image/	HO, BG, HS	2
create folder for masked images && masks (w/o images) in ROI script	HO, BG, HS	2
scale masks to original resolution of the image for later use	HO, BG, HS	3
organize the GitHub repo (remove uneeded scripts and standardize naming conventions, etc.)	HS	3
matplotlib and interactive matplotlib issues on Mac OS	HS, BG, HO	3
more prominent and better date show on each image slide	HO	3
prepare presentation for 6/21	HS, BG, HO	Other
figure out our file hosting options: server, google drive, local files, etc.	AT, WS	3
finish minimum viable product (Bookdown) first draft, setup Github pages	HS, HO	3
edit bookdown to support generalized usage	HS, HO, BG	4
familiarize with ML algorithm used	HO, HS, BG	5
train and test random forest classifier	HO, HS, BG	5

Future Objectives

Tasks	People	Step
create walkthrough with example files as both proof-of-concept for our data pipeline, and as a walkthrough for future users	XX	7
export Jupyter notebook to .py files	XX	8

Notes

While the above steps are roughly in order, we will be working on certain aspects of the project throughout, such as:

• Keeping all code well-documented and clean
• Creating Jupyter notebooks for each script with embedded markdown/html that explains each step of the script
• Determining a final product to make data pipeline easily accessible to average scientist

Progress

• Figured out how to click through image files with mask applied
• Figured outredrawing ROI and applying new mask to remaining images
• Working on file naming conventions/organizing GitHub

Repo Structure

README.md

data

• derived
• munged
• raw

scripts

• 01-folder
• 02-folder
• etc.
• dep (dependencies)
• sandbox

bookdown

• index.rmd
• 01.rmd
• .yml files

.gitignore

.Rproj