Open andrewsg3 opened 3 years ago
Current approach is a HAAR cascade trained on sample images that have had a green HSV mask and findContours() passed over them. It obtained one cascade with a low level of false negatives but that only grabs a handful of seedlings in the scene. I am also finding an issue that the cascade trainer GUI runs out of memory at samples of P>250 and N>350, which is effectively setting a limit on how extensively I can train it. This is with 8192mb ram assigned.
I will try: -Training on green masked images without findContours() passed over them -Selecting what I consider the best samples out of all available for the training dataset given the limitations
Is your feature request related to a problem? Please describe. To be able to recognise the presence of soil mold from a still image or camera feed
Describe the solution you'd like Python implementation based around 'Lazy tech' method https://www.youtube.com/watch?v=_gGI91BmIdk&list=PLAs-3cqyNbIjGzf50LckxBndLCd1EgB0w&index=3
Describe alternatives you've considered
Additional context