Check out https://github.com/thtrieu/darkflow
This repo is an in-development research effort to count ladies slipper orchids from drone footage for the purpose of monitoring populations.
Open up a terminal, cd to this directory, run:
make
contains validation images for run.py
yolo weights
These can be found (link somewhere) on the Darkflow GitHub Add the downloaded weights to 'bin' and rename to 'yolo.weights'
Add the image to test to 'sample_img' Modify 'run.py' to use the new images 'image_name'
python3 run.py
prints anchors to use in config
train the counter
test using image
test using video (WIP)
add class names here, each on a new line
update 'cfg_classes'
note down details to use next
under [region], adjust 'num' and 'anchors', adjust 'filters' from convolutional layer above
adjust (at top) 'width' and 'height' of image to multiples of 32
match 'cfg_width' and 'cfg_height'