Home

Done

TODO

---

tech

• ! test new install - fix preinstall on newly downloaded repo

• test tslint ok

• test bash scripts on Linux - ls -l for Mac

• ? consider moving to shx - shell for nodeJs: https://www.npmjs.com/package/shx

• ? state/result cache? consider using sqlLite for nodeJs: https://www.npmjs.com/package/websql

• add dead code tslint - also in tslint ban, tslint-folders https://eslint.org/docs/rules/no-unused-vars

• ? try reduce install size (pkg size is ok) - rollup can bundle all except Google, sharp

• try make main.ts import from the bundle file - then can make all 'dev' except Google, sharp

• ? merge greenkeeper branches, re-yarn locally

• ? reduce dependencies by relying on rollup - but could be fragile, hard to test... - move to webpack?

• (BREAKING CHANGE) use same ENV var for the Vision API as the GeoCoding - then is simpler to setup

• hide/move detailed output -verbose via IOutputter

• n/a to logfile - log4net like auto new files per day

• n/a - quality is not as good and still only JPEG (ref: chocolate-bars) try move to exif-reader as is MIT - see https://www.npmjs.com/package/exif-reader

• replace ExifReader with another that is MIT - https://www.npmjs.com/package/exif-parser

• ? try yargs parser

• ? try snyk - vulnerability scanner https://snyk.io/test/github/inversify/InversifyJS

• ? use moment.js for the 'next date', 'date range' handling ? (CSV)

• ? fork the starter kit for future use

---

unit tests

• add more unit tests

• unit test getting props of dir images (the API used as a library)

• mock out the moving

• mock out the vision API

• mock out the geo API

• re-enable the code coverage in package.json - see coveragePathIgnorePatterns - 80% good enough

• add back the coveralls badge (see commit of 5 Dec 2018)

---

features

• fix: in dryRun, the classifyAndMove phase always returns result count of 0 (important for use as library)

• add -geocodeOnly option, so no phase 2

• add -parallel for performance - parallelize calls to geo, vision api

• add image format - PNG

• add image format - TIFF (include exif) - try to keep 'ColorSpace' equivalent (so can see sRGB)

• ? expose resize threshold in MB - cli + lib

---

image quality filter

• separate project! re-use here

[fail] - try poor quality images for labels:

• Google Vision does not support - only negative label seen is 'haze'

• use crop suggestions (or regions) from Google Vision to focus the quality check

• detect out of focus (blurred) images

• detect over exposed

• detect under exposed

• low contrast

• face detection - face out of frame, not horizontal

(see crafting)

---

add more Vision API results

• imageProperties
• landmarkDetection
• logoDetection
• ? face detection

---

future

• ? combine labels into 'well known' labels e.g. building, architecture -> architecture